tag:blogger.com,1999:blog-198032222024-03-18T01:45:45.784-06:00natural language processing blogmy biased thoughts on the fields of natural language processing (NLP), computational linguistics (CL) and related topics (machine learning, math, funding, etc.)halhttp://www.blogger.com/profile/02162908373916390369noreply@blogger.comBlogger319125tag:blogger.com,1999:blog-19803222.post-1185701030441395872018-07-16T14:26:00.003-06:002018-07-16T14:26:54.921-06:00Yet another list of things we can do to have more diverse sets of invited speakersI am super reticent to write this because <i>so many people</i> have written similar things. Yet despite that, we still have things like an initial workshop program with 17 invited speakers of whom 16 are men (at ICML 2018) and a panel where 3 of the 5 panelists are or were previously in the same group at Stanford (at NAACL 2018). And I (and others) still hear things like "tell me what to do." So, here it is, I'm saying what we should do. (And I'm saying this mostly so that the others, who are usually women, don't have to keep dealing with this and can hopefully point here.)<br />
<br />
Like any such list, this is incomplete. I'd love suggestions for improvements. I'm also taking for granted that you <i>want</i> a diverse set of invited speakers. If you don't, go read up on that first.<br />
<br />
I think it's also important to recognize that we all mess up. I (Hal) have messed up many times. The important thing is to learn. In order to learn, we need a signal, and so if someone points out that your event did not live up to what-should-be-expectations, accept that. Don't get defensive. Realize that this learning signal is super rare, and so apply some <a href="https://en.wikipedia.org/wiki/Inverse_probability_weighting">inverse propensity weighting</a> and do a big learning update.<br />
<br />
This list was created by my first brainstorming ideas (largely based on my own failures), then going to some excellent sources and adding/modifying. You should also read these sources (which mostly focus on panels, not invited speakers, and also mostly focus on gender):<br />
<ul>
<li> <a href="https://www.theguardian.com/commentisfree/2013/nov/06/four-steps-to-put-an-end-to-all-male-panels-at-conferences">Four steps to put an end to all-male panels at conferences</a>, Bronwen Clune, The Guardian, 2013 </li>
<li> <a href="https://www.huffingtonpost.com/entry/all-male-panel-pledge_us_56e848f9e4b065e2e3d77148">Stop Agreeing To Be On All-Male Panels Just Stop</a>, Emily Peck, Huffington Post, 2016 </li>
<li> <a href="https://onthinktanks.org/articles/there-is-no-excuse-for-an-all-male-panel-how-to-avoid-them/">How to avoid an all-male panel? Just try harder</a>, Enrique Mendizabal, On Think Tanks, 2016 </li>
<li> <a href="http://foreignpolicy.com/2016/03/08/7-rules-for-avoiding-all-male-panels/">7 Rules for Avoiding All-Male Panels</a>, Jacqueline O'Neill, Foreign Policy, 2016 </li>
<li> <a href="http://www.slate.com/blogs/better_life_lab/2017/10/06/there_is_no_excuse_for_all_male_panels_here_s_how_to_fix_them.html">There’s No Excuse for All-Male Panels. Here’s How to Fix Them</a>, Brigid Schulte, Slate, 2017 </li>
<li> <a href="https://www.replyall.me/jofas-cast/preventing-all-male-panels/">Preventing All Male Panels</a>, Sharon Weiss-Greenberg and Panelists, ReplyAll </li>
</ul>
Okay, so here's my list, broken down into "phases" of the workshop organization process.<br />
<ul>
<li> <b> Before you begin inviting folks </b>
<ul>
<li> <span style="color: #38761d;"><b>Have an goal</b></span>, write it down, and routinely evaluate yourself against that goal. If you succeed, great. If you fail, learn (also great!). </li>
<li><span style="color: #38761d;"><b>Have co-organizers with different social networks</b></span> than you have, or you'll all only think of the same people. </li>
<li><span style="color: #38761d;"><b> Start with an initial diverse list</b></span> of potential speakers that's <i>mostly</i> (well more than half) women and/or minorities, covering different geographic regions, different universities, and different levels of "seniority". You need to start with well more than half because (a) you should expect many to say no, and because (b) many of your contributed papers are highly likely presented by abled white guys from privileged institutions in the US, so if the event is to be even remotely balanced you need to compensate early. </li>
<li> <span style="color: #38761d;"><b>Scan through the proceedings </b></span>of recent conferences for people that aren't immediately on your radar. </li>
<li> If you can't come up with a long enough list that's also diverse, then maybe <span style="color: #38761d;"><b>consider whether your topic is just "you and your buddies,"</b></span> and perhaps think about if you can expand your topic in an interesting direction to cast a broader net. </li>
<li> If you can't come up with such a list, <span style="color: #38761d;"><b>maybe your criteria for who to invite is unrealistic</b></span> already very white male-biased. For instance, having a criteria like "I only want full profs from US universities" comes with a lot of historical/social baggage. </li>
<li><span style="color: #38761d;"><b> Ask everyone you know for suggested names</b></span>. Check out existing resources like the <a href="http://wimlworkshop.org/directory-of-women-in-machine-learning/">WiML</a> and <a href="http://www.winlp.org/big-directory/">Widening NLP</a> directories, but also realize that there are many forms of diversity that may not be well covered in these. </li>
<li> Once you have long list of potential speakers with many women or minorities on it, <span style="color: #38761d;"><b>ensure that you're not just inviting women to talk about "soft" topics</b></span> and men about "technical" topics. </li>
</ul>
</li>
<li> <b> In the invitation process </b>
<ul>
<li> In the invitation letter to speakers, <span style="color: #38761d;"><b>offer to cover childcare</b></span> for the speaker (regardless of who it is) either at the workshop or at their home city. Women in particular often take the majority of child rearing responsibilities; this may help tip the scales, but will also help everyone who has kids. </li>
<li> In each invitation that you send out to men, or people who are not under-represented, <span style="color: #38761d;"><b>ask them <i>explicitly</i> for suggestions of additional speakers</b></span> (who are not white men) you could invite in the initial invitation (i.e., not just when they decline). </li>
<li> <span style="color: #38761d;"><b>Invite speakers from under-represented or historically excluded groups <i>very early</i> </b></span>before they become even more overcommitted. But also give them an easy out to say no. </li>
<li> When you start sending invitations out, <span style="color: #38761d;"><b>invite the abled white guys at </b></span><span style="color: #38761d;"><b><span style="color: #38761d;"><b>privileged institutions </b></span>slowly</b></span> and later. That way, if you have trouble getting a diverse set of speakers, you’re not already overcommitted. </li>
</ul>
</li>
<li> <b> Dealing with challenges </b>
<ul>
<li> If the diversity of your event is being hurt by the fact that potential speakers cannot travel, <span style="color: #38761d;"><b>consider allowing one or two people speak remotely</b></span>. </li>
<li> If you do find yourself overcommitted to a non-diverse speaker group, <span style="color: #38761d;"><b>it may be time to eat crow, apologize</b></span> to a few of them, and say directly that you were aiming for a diverse slate of speakers, but you messed up, and you would like to know if they would be willing to step down to allow room for someone else to speak in their stead. </li>
<li> <span style="color: #38761d;"><b>Go back to your goals.</b></span> How are you doing? What can you do better next time?</li>
</ul>
</li>
</ul>
<br />
Finally, if you're a guy, or otherwise hold significant privilege, even if you're not organizing a workshop try to help people
who are. You should have a go-to list of alternate speakers that you can
provide when colleagues ask you for ideas of who to invite or when you get invited yourself. You can have
an <a href="https://en.wikipedia.org/wiki/Inclusion_rider">inclusion</a> <a href="http://hal3.name/faq.html">rider</a> for giving talks and being on panels, and perhaps also for putting your name on workshops as a co-organizer. I promise, people will appreciate the help!<br />
<br />halhttp://www.blogger.com/profile/02162908373916390369noreply@blogger.com11tag:blogger.com,1999:blog-19803222.post-58956899302106480732018-06-12T12:22:00.000-06:002018-06-13T07:58:55.582-06:00Many opportunities for discrimination in deploying machine learning systems<!--[if gte mso 9]><xml>
<o:OfficeDocumentSettings>
<o:AllowPNG/>
</o:OfficeDocumentSettings>
</xml><![endif]--><!--[if gte mso 9]><xml>
<w:WordDocument>
<w:View>Normal</w:View>
<w:Zoom>0</w:Zoom>
<w:TrackMoves/>
<w:TrackFormatting/>
<w:PunctuationKerning/>
<w:ValidateAgainstSchemas/>
<w:SaveIfXMLInvalid>false</w:SaveIfXMLInvalid>
<w:IgnoreMixedContent>false</w:IgnoreMixedContent>
<w:AlwaysShowPlaceholderText>false</w:AlwaysShowPlaceholderText>
<w:DoNotPromoteQF/>
<w:LidThemeOther>EN-US</w:LidThemeOther>
<w:LidThemeAsian>X-NONE</w:LidThemeAsian>
<w:LidThemeComplexScript>X-NONE</w:LidThemeComplexScript>
<w:Compatibility>
<w:BreakWrappedTables/>
<w:SnapToGridInCell/>
<w:WrapTextWithPunct/>
<w:UseAsianBreakRules/>
<w:DontGrowAutofit/>
<w:SplitPgBreakAndParaMark/>
<w:EnableOpenTypeKerning/>
<w:DontFlipMirrorIndents/>
<w:OverrideTableStyleHps/>
</w:Compatibility>
<m:mathPr>
<m:mathFont m:val="Cambria Math"/>
<m:brkBin m:val="before"/>
<m:brkBinSub m:val="--"/>
<m:smallFrac m:val="off"/>
<m:dispDef/>
<m:lMargin m:val="0"/>
<m:rMargin m:val="0"/>
<m:defJc m:val="centerGroup"/>
<m:wrapIndent m:val="1440"/>
<m:intLim m:val="subSup"/>
<m:naryLim m:val="undOvr"/>
</m:mathPr></w:WordDocument>
</xml><![endif]--><!--[if gte mso 9]><xml>
<w:LatentStyles DefLockedState="false" DefUnhideWhenUsed="false"
DefSemiHidden="false" DefQFormat="false" DefPriority="99"
LatentStyleCount="375">
<w:LsdException Locked="false" Priority="0" QFormat="true" Name="Normal"/>
<w:LsdException Locked="false" Priority="9" QFormat="true" Name="heading 1"/>
<w:LsdException Locked="false" Priority="9" SemiHidden="true"
UnhideWhenUsed="true" QFormat="true" Name="heading 2"/>
<w:LsdException Locked="false" Priority="9" SemiHidden="true"
UnhideWhenUsed="true" QFormat="true" Name="heading 3"/>
<w:LsdException Locked="false" Priority="9" SemiHidden="true"
UnhideWhenUsed="true" QFormat="true" Name="heading 4"/>
<w:LsdException Locked="false" Priority="9" SemiHidden="true"
UnhideWhenUsed="true" QFormat="true" Name="heading 5"/>
<w:LsdException Locked="false" Priority="9" SemiHidden="true"
UnhideWhenUsed="true" QFormat="true" Name="heading 6"/>
<w:LsdException Locked="false" Priority="9" SemiHidden="true"
UnhideWhenUsed="true" QFormat="true" Name="heading 7"/>
<w:LsdException Locked="false" Priority="9" SemiHidden="true"
UnhideWhenUsed="true" QFormat="true" Name="heading 8"/>
<w:LsdException Locked="false" Priority="9" SemiHidden="true"
UnhideWhenUsed="true" QFormat="true" Name="heading 9"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="index 1"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="index 2"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="index 3"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="index 4"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="index 5"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="index 6"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="index 7"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="index 8"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="index 9"/>
<w:LsdException Locked="false" Priority="39" SemiHidden="true"
UnhideWhenUsed="true" Name="toc 1"/>
<w:LsdException Locked="false" Priority="39" SemiHidden="true"
UnhideWhenUsed="true" Name="toc 2"/>
<w:LsdException Locked="false" Priority="39" SemiHidden="true"
UnhideWhenUsed="true" Name="toc 3"/>
<w:LsdException Locked="false" Priority="39" SemiHidden="true"
UnhideWhenUsed="true" Name="toc 4"/>
<w:LsdException Locked="false" Priority="39" SemiHidden="true"
UnhideWhenUsed="true" Name="toc 5"/>
<w:LsdException Locked="false" Priority="39" SemiHidden="true"
UnhideWhenUsed="true" Name="toc 6"/>
<w:LsdException Locked="false" Priority="39" SemiHidden="true"
UnhideWhenUsed="true" Name="toc 7"/>
<w:LsdException Locked="false" Priority="39" SemiHidden="true"
UnhideWhenUsed="true" Name="toc 8"/>
<w:LsdException Locked="false" Priority="39" SemiHidden="true"
UnhideWhenUsed="true" Name="toc 9"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="Normal Indent"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="footnote text"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="annotation text"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="header"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="footer"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="index heading"/>
<w:LsdException Locked="false" Priority="35" SemiHidden="true"
UnhideWhenUsed="true" QFormat="true" Name="caption"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="table of figures"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="envelope address"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="envelope return"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="footnote reference"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="annotation reference"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="line number"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="page number"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="endnote reference"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="endnote text"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="table of authorities"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="macro"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="toa heading"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="List"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="List Bullet"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="List Number"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="List 2"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="List 3"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="List 4"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="List 5"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="List Bullet 2"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="List Bullet 3"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="List Bullet 4"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="List Bullet 5"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="List Number 2"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="List Number 3"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="List Number 4"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="List Number 5"/>
<w:LsdException Locked="false" Priority="10" QFormat="true" Name="Title"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="Closing"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="Signature"/>
<w:LsdException Locked="false" Priority="1" SemiHidden="true"
UnhideWhenUsed="true" Name="Default Paragraph Font"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="Body Text"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="Body Text Indent"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="List Continue"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="List Continue 2"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="List Continue 3"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="List Continue 4"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="List Continue 5"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="Message Header"/>
<w:LsdException Locked="false" Priority="11" QFormat="true" Name="Subtitle"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="Salutation"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="Date"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="Body Text First Indent"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="Body Text First Indent 2"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="Note Heading"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="Body Text 2"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="Body Text 3"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="Body Text Indent 2"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="Body Text Indent 3"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="Block Text"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="Hyperlink"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="FollowedHyperlink"/>
<w:LsdException Locked="false" Priority="22" QFormat="true" Name="Strong"/>
<w:LsdException Locked="false" Priority="20" QFormat="true" Name="Emphasis"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="Document Map"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="Plain Text"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="E-mail Signature"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="HTML Top of Form"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="HTML Bottom of Form"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="Normal (Web)"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="HTML Acronym"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="HTML Address"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="HTML Cite"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="HTML Code"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="HTML Definition"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="HTML Keyboard"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="HTML Preformatted"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="HTML Sample"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="HTML Typewriter"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="HTML Variable"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="Normal Table"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="annotation subject"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="No List"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="Outline List 1"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="Outline List 2"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="Outline List 3"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="Table Simple 1"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="Table Simple 2"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="Table Simple 3"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="Table Classic 1"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="Table Classic 2"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="Table Classic 3"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="Table Classic 4"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="Table Colorful 1"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="Table Colorful 2"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="Table Colorful 3"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="Table Columns 1"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="Table Columns 2"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="Table Columns 3"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="Table Columns 4"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="Table Columns 5"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="Table Grid 1"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="Table Grid 2"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="Table Grid 3"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="Table Grid 4"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="Table Grid 5"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="Table Grid 6"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="Table Grid 7"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="Table Grid 8"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="Table List 1"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="Table List 2"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="Table List 3"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="Table List 4"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="Table List 5"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="Table List 6"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="Table List 7"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="Table List 8"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="Table 3D effects 1"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="Table 3D effects 2"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="Table 3D effects 3"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="Table Contemporary"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="Table Elegant"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="Table Professional"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="Table Subtle 1"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="Table Subtle 2"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="Table Web 1"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="Table Web 2"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="Table Web 3"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="Balloon Text"/>
<w:LsdException Locked="false" Priority="39" Name="Table Grid"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="Table Theme"/>
<w:LsdException Locked="false" SemiHidden="true" Name="Placeholder Text"/>
<w:LsdException Locked="false" Priority="1" QFormat="true" Name="No Spacing"/>
<w:LsdException Locked="false" Priority="60" Name="Light Shading"/>
<w:LsdException Locked="false" Priority="61" Name="Light List"/>
<w:LsdException Locked="false" Priority="62" Name="Light Grid"/>
<w:LsdException Locked="false" Priority="63" Name="Medium Shading 1"/>
<w:LsdException Locked="false" Priority="64" Name="Medium Shading 2"/>
<w:LsdException Locked="false" Priority="65" Name="Medium List 1"/>
<w:LsdException Locked="false" Priority="66" Name="Medium List 2"/>
<w:LsdException Locked="false" Priority="67" Name="Medium Grid 1"/>
<w:LsdException Locked="false" Priority="68" Name="Medium Grid 2"/>
<w:LsdException Locked="false" Priority="69" Name="Medium Grid 3"/>
<w:LsdException Locked="false" Priority="70" Name="Dark List"/>
<w:LsdException Locked="false" Priority="71" Name="Colorful Shading"/>
<w:LsdException Locked="false" Priority="72" Name="Colorful List"/>
<w:LsdException Locked="false" Priority="73" Name="Colorful Grid"/>
<w:LsdException Locked="false" Priority="60" Name="Light Shading Accent 1"/>
<w:LsdException Locked="false" Priority="61" Name="Light List Accent 1"/>
<w:LsdException Locked="false" Priority="62" Name="Light Grid Accent 1"/>
<w:LsdException Locked="false" Priority="63" Name="Medium Shading 1 Accent 1"/>
<w:LsdException Locked="false" Priority="64" Name="Medium Shading 2 Accent 1"/>
<w:LsdException Locked="false" Priority="65" Name="Medium List 1 Accent 1"/>
<w:LsdException Locked="false" SemiHidden="true" Name="Revision"/>
<w:LsdException Locked="false" Priority="34" QFormat="true"
Name="List Paragraph"/>
<w:LsdException Locked="false" Priority="29" QFormat="true" Name="Quote"/>
<w:LsdException Locked="false" Priority="30" QFormat="true"
Name="Intense Quote"/>
<w:LsdException Locked="false" Priority="66" Name="Medium List 2 Accent 1"/>
<w:LsdException Locked="false" Priority="67" Name="Medium Grid 1 Accent 1"/>
<w:LsdException Locked="false" Priority="68" Name="Medium Grid 2 Accent 1"/>
<w:LsdException Locked="false" Priority="69" Name="Medium Grid 3 Accent 1"/>
<w:LsdException Locked="false" Priority="70" Name="Dark List Accent 1"/>
<w:LsdException Locked="false" Priority="71" Name="Colorful Shading Accent 1"/>
<w:LsdException Locked="false" Priority="72" Name="Colorful List Accent 1"/>
<w:LsdException Locked="false" Priority="73" Name="Colorful Grid Accent 1"/>
<w:LsdException Locked="false" Priority="60" Name="Light Shading Accent 2"/>
<w:LsdException Locked="false" Priority="61" Name="Light List Accent 2"/>
<w:LsdException Locked="false" Priority="62" Name="Light Grid Accent 2"/>
<w:LsdException Locked="false" Priority="63" Name="Medium Shading 1 Accent 2"/>
<w:LsdException Locked="false" Priority="64" Name="Medium Shading 2 Accent 2"/>
<w:LsdException Locked="false" Priority="65" Name="Medium List 1 Accent 2"/>
<w:LsdException Locked="false" Priority="66" Name="Medium List 2 Accent 2"/>
<w:LsdException Locked="false" Priority="67" Name="Medium Grid 1 Accent 2"/>
<w:LsdException Locked="false" Priority="68" Name="Medium Grid 2 Accent 2"/>
<w:LsdException Locked="false" Priority="69" Name="Medium Grid 3 Accent 2"/>
<w:LsdException Locked="false" Priority="70" Name="Dark List Accent 2"/>
<w:LsdException Locked="false" Priority="71" Name="Colorful Shading Accent 2"/>
<w:LsdException Locked="false" Priority="72" Name="Colorful List Accent 2"/>
<w:LsdException Locked="false" Priority="73" Name="Colorful Grid Accent 2"/>
<w:LsdException Locked="false" Priority="60" Name="Light Shading Accent 3"/>
<w:LsdException Locked="false" Priority="61" Name="Light List Accent 3"/>
<w:LsdException Locked="false" Priority="62" Name="Light Grid Accent 3"/>
<w:LsdException Locked="false" Priority="63" Name="Medium Shading 1 Accent 3"/>
<w:LsdException Locked="false" Priority="64" Name="Medium Shading 2 Accent 3"/>
<w:LsdException Locked="false" Priority="65" Name="Medium List 1 Accent 3"/>
<w:LsdException Locked="false" Priority="66" Name="Medium List 2 Accent 3"/>
<w:LsdException Locked="false" Priority="67" Name="Medium Grid 1 Accent 3"/>
<w:LsdException Locked="false" Priority="68" Name="Medium Grid 2 Accent 3"/>
<w:LsdException Locked="false" Priority="69" Name="Medium Grid 3 Accent 3"/>
<w:LsdException Locked="false" Priority="70" Name="Dark List Accent 3"/>
<w:LsdException Locked="false" Priority="71" Name="Colorful Shading Accent 3"/>
<w:LsdException Locked="false" Priority="72" Name="Colorful List Accent 3"/>
<w:LsdException Locked="false" Priority="73" Name="Colorful Grid Accent 3"/>
<w:LsdException Locked="false" Priority="60" Name="Light Shading Accent 4"/>
<w:LsdException Locked="false" Priority="61" Name="Light List Accent 4"/>
<w:LsdException Locked="false" Priority="62" Name="Light Grid Accent 4"/>
<w:LsdException Locked="false" Priority="63" Name="Medium Shading 1 Accent 4"/>
<w:LsdException Locked="false" Priority="64" Name="Medium Shading 2 Accent 4"/>
<w:LsdException Locked="false" Priority="65" Name="Medium List 1 Accent 4"/>
<w:LsdException Locked="false" Priority="66" Name="Medium List 2 Accent 4"/>
<w:LsdException Locked="false" Priority="67" Name="Medium Grid 1 Accent 4"/>
<w:LsdException Locked="false" Priority="68" Name="Medium Grid 2 Accent 4"/>
<w:LsdException Locked="false" Priority="69" Name="Medium Grid 3 Accent 4"/>
<w:LsdException Locked="false" Priority="70" Name="Dark List Accent 4"/>
<w:LsdException Locked="false" Priority="71" Name="Colorful Shading Accent 4"/>
<w:LsdException Locked="false" Priority="72" Name="Colorful List Accent 4"/>
<w:LsdException Locked="false" Priority="73" Name="Colorful Grid Accent 4"/>
<w:LsdException Locked="false" Priority="60" Name="Light Shading Accent 5"/>
<w:LsdException Locked="false" Priority="61" Name="Light List Accent 5"/>
<w:LsdException Locked="false" Priority="62" Name="Light Grid Accent 5"/>
<w:LsdException Locked="false" Priority="63" Name="Medium Shading 1 Accent 5"/>
<w:LsdException Locked="false" Priority="64" Name="Medium Shading 2 Accent 5"/>
<w:LsdException Locked="false" Priority="65" Name="Medium List 1 Accent 5"/>
<w:LsdException Locked="false" Priority="66" Name="Medium List 2 Accent 5"/>
<w:LsdException Locked="false" Priority="67" Name="Medium Grid 1 Accent 5"/>
<w:LsdException Locked="false" Priority="68" Name="Medium Grid 2 Accent 5"/>
<w:LsdException Locked="false" Priority="69" Name="Medium Grid 3 Accent 5"/>
<w:LsdException Locked="false" Priority="70" Name="Dark List Accent 5"/>
<w:LsdException Locked="false" Priority="71" Name="Colorful Shading Accent 5"/>
<w:LsdException Locked="false" Priority="72" Name="Colorful List Accent 5"/>
<w:LsdException Locked="false" Priority="73" Name="Colorful Grid Accent 5"/>
<w:LsdException Locked="false" Priority="60" Name="Light Shading Accent 6"/>
<w:LsdException Locked="false" Priority="61" Name="Light List Accent 6"/>
<w:LsdException Locked="false" Priority="62" Name="Light Grid Accent 6"/>
<w:LsdException Locked="false" Priority="63" Name="Medium Shading 1 Accent 6"/>
<w:LsdException Locked="false" Priority="64" Name="Medium Shading 2 Accent 6"/>
<w:LsdException Locked="false" Priority="65" Name="Medium List 1 Accent 6"/>
<w:LsdException Locked="false" Priority="66" Name="Medium List 2 Accent 6"/>
<w:LsdException Locked="false" Priority="67" Name="Medium Grid 1 Accent 6"/>
<w:LsdException Locked="false" Priority="68" Name="Medium Grid 2 Accent 6"/>
<w:LsdException Locked="false" Priority="69" Name="Medium Grid 3 Accent 6"/>
<w:LsdException Locked="false" Priority="70" Name="Dark List Accent 6"/>
<w:LsdException Locked="false" Priority="71" Name="Colorful Shading Accent 6"/>
<w:LsdException Locked="false" Priority="72" Name="Colorful List Accent 6"/>
<w:LsdException Locked="false" Priority="73" Name="Colorful Grid Accent 6"/>
<w:LsdException Locked="false" Priority="19" QFormat="true"
Name="Subtle Emphasis"/>
<w:LsdException Locked="false" Priority="21" QFormat="true"
Name="Intense Emphasis"/>
<w:LsdException Locked="false" Priority="31" QFormat="true"
Name="Subtle Reference"/>
<w:LsdException Locked="false" Priority="32" QFormat="true"
Name="Intense Reference"/>
<w:LsdException Locked="false" Priority="33" QFormat="true" Name="Book Title"/>
<w:LsdException Locked="false" Priority="37" SemiHidden="true"
UnhideWhenUsed="true" Name="Bibliography"/>
<w:LsdException Locked="false" Priority="39" SemiHidden="true"
UnhideWhenUsed="true" QFormat="true" Name="TOC Heading"/>
<w:LsdException Locked="false" Priority="41" Name="Plain Table 1"/>
<w:LsdException Locked="false" Priority="42" Name="Plain Table 2"/>
<w:LsdException Locked="false" Priority="43" Name="Plain Table 3"/>
<w:LsdException Locked="false" Priority="44" Name="Plain Table 4"/>
<w:LsdException Locked="false" Priority="45" Name="Plain Table 5"/>
<w:LsdException Locked="false" Priority="40" Name="Grid Table Light"/>
<w:LsdException Locked="false" Priority="46" Name="Grid Table 1 Light"/>
<w:LsdException Locked="false" Priority="47" Name="Grid Table 2"/>
<w:LsdException Locked="false" Priority="48" Name="Grid Table 3"/>
<w:LsdException Locked="false" Priority="49" Name="Grid Table 4"/>
<w:LsdException Locked="false" Priority="50" Name="Grid Table 5 Dark"/>
<w:LsdException Locked="false" Priority="51" Name="Grid Table 6 Colorful"/>
<w:LsdException Locked="false" Priority="52" Name="Grid Table 7 Colorful"/>
<w:LsdException Locked="false" Priority="46"
Name="Grid Table 1 Light Accent 1"/>
<w:LsdException Locked="false" Priority="47" Name="Grid Table 2 Accent 1"/>
<w:LsdException Locked="false" Priority="48" Name="Grid Table 3 Accent 1"/>
<w:LsdException Locked="false" Priority="49" Name="Grid Table 4 Accent 1"/>
<w:LsdException Locked="false" Priority="50" Name="Grid Table 5 Dark Accent 1"/>
<w:LsdException Locked="false" Priority="51"
Name="Grid Table 6 Colorful Accent 1"/>
<w:LsdException Locked="false" Priority="52"
Name="Grid Table 7 Colorful Accent 1"/>
<w:LsdException Locked="false" Priority="46"
Name="Grid Table 1 Light Accent 2"/>
<w:LsdException Locked="false" Priority="47" Name="Grid Table 2 Accent 2"/>
<w:LsdException Locked="false" Priority="48" Name="Grid Table 3 Accent 2"/>
<w:LsdException Locked="false" Priority="49" Name="Grid Table 4 Accent 2"/>
<w:LsdException Locked="false" Priority="50" Name="Grid Table 5 Dark Accent 2"/>
<w:LsdException Locked="false" Priority="51"
Name="Grid Table 6 Colorful Accent 2"/>
<w:LsdException Locked="false" Priority="52"
Name="Grid Table 7 Colorful Accent 2"/>
<w:LsdException Locked="false" Priority="46"
Name="Grid Table 1 Light Accent 3"/>
<w:LsdException Locked="false" Priority="47" Name="Grid Table 2 Accent 3"/>
<w:LsdException Locked="false" Priority="48" Name="Grid Table 3 Accent 3"/>
<w:LsdException Locked="false" Priority="49" Name="Grid Table 4 Accent 3"/>
<w:LsdException Locked="false" Priority="50" Name="Grid Table 5 Dark Accent 3"/>
<w:LsdException Locked="false" Priority="51"
Name="Grid Table 6 Colorful Accent 3"/>
<w:LsdException Locked="false" Priority="52"
Name="Grid Table 7 Colorful Accent 3"/>
<w:LsdException Locked="false" Priority="46"
Name="Grid Table 1 Light Accent 4"/>
<w:LsdException Locked="false" Priority="47" Name="Grid Table 2 Accent 4"/>
<w:LsdException Locked="false" Priority="48" Name="Grid Table 3 Accent 4"/>
<w:LsdException Locked="false" Priority="49" Name="Grid Table 4 Accent 4"/>
<w:LsdException Locked="false" Priority="50" Name="Grid Table 5 Dark Accent 4"/>
<w:LsdException Locked="false" Priority="51"
Name="Grid Table 6 Colorful Accent 4"/>
<w:LsdException Locked="false" Priority="52"
Name="Grid Table 7 Colorful Accent 4"/>
<w:LsdException Locked="false" Priority="46"
Name="Grid Table 1 Light Accent 5"/>
<w:LsdException Locked="false" Priority="47" Name="Grid Table 2 Accent 5"/>
<w:LsdException Locked="false" Priority="48" Name="Grid Table 3 Accent 5"/>
<w:LsdException Locked="false" Priority="49" Name="Grid Table 4 Accent 5"/>
<w:LsdException Locked="false" Priority="50" Name="Grid Table 5 Dark Accent 5"/>
<w:LsdException Locked="false" Priority="51"
Name="Grid Table 6 Colorful Accent 5"/>
<w:LsdException Locked="false" Priority="52"
Name="Grid Table 7 Colorful Accent 5"/>
<w:LsdException Locked="false" Priority="46"
Name="Grid Table 1 Light Accent 6"/>
<w:LsdException Locked="false" Priority="47" Name="Grid Table 2 Accent 6"/>
<w:LsdException Locked="false" Priority="48" Name="Grid Table 3 Accent 6"/>
<w:LsdException Locked="false" Priority="49" Name="Grid Table 4 Accent 6"/>
<w:LsdException Locked="false" Priority="50" Name="Grid Table 5 Dark Accent 6"/>
<w:LsdException Locked="false" Priority="51"
Name="Grid Table 6 Colorful Accent 6"/>
<w:LsdException Locked="false" Priority="52"
Name="Grid Table 7 Colorful Accent 6"/>
<w:LsdException Locked="false" Priority="46" Name="List Table 1 Light"/>
<w:LsdException Locked="false" Priority="47" Name="List Table 2"/>
<w:LsdException Locked="false" Priority="48" Name="List Table 3"/>
<w:LsdException Locked="false" Priority="49" Name="List Table 4"/>
<w:LsdException Locked="false" Priority="50" Name="List Table 5 Dark"/>
<w:LsdException Locked="false" Priority="51" Name="List Table 6 Colorful"/>
<w:LsdException Locked="false" Priority="52" Name="List Table 7 Colorful"/>
<w:LsdException Locked="false" Priority="46"
Name="List Table 1 Light Accent 1"/>
<w:LsdException Locked="false" Priority="47" Name="List Table 2 Accent 1"/>
<w:LsdException Locked="false" Priority="48" Name="List Table 3 Accent 1"/>
<w:LsdException Locked="false" Priority="49" Name="List Table 4 Accent 1"/>
<w:LsdException Locked="false" Priority="50" Name="List Table 5 Dark Accent 1"/>
<w:LsdException Locked="false" Priority="51"
Name="List Table 6 Colorful Accent 1"/>
<w:LsdException Locked="false" Priority="52"
Name="List Table 7 Colorful Accent 1"/>
<w:LsdException Locked="false" Priority="46"
Name="List Table 1 Light Accent 2"/>
<w:LsdException Locked="false" Priority="47" Name="List Table 2 Accent 2"/>
<w:LsdException Locked="false" Priority="48" Name="List Table 3 Accent 2"/>
<w:LsdException Locked="false" Priority="49" Name="List Table 4 Accent 2"/>
<w:LsdException Locked="false" Priority="50" Name="List Table 5 Dark Accent 2"/>
<w:LsdException Locked="false" Priority="51"
Name="List Table 6 Colorful Accent 2"/>
<w:LsdException Locked="false" Priority="52"
Name="List Table 7 Colorful Accent 2"/>
<w:LsdException Locked="false" Priority="46"
Name="List Table 1 Light Accent 3"/>
<w:LsdException Locked="false" Priority="47" Name="List Table 2 Accent 3"/>
<w:LsdException Locked="false" Priority="48" Name="List Table 3 Accent 3"/>
<w:LsdException Locked="false" Priority="49" Name="List Table 4 Accent 3"/>
<w:LsdException Locked="false" Priority="50" Name="List Table 5 Dark Accent 3"/>
<w:LsdException Locked="false" Priority="51"
Name="List Table 6 Colorful Accent 3"/>
<w:LsdException Locked="false" Priority="52"
Name="List Table 7 Colorful Accent 3"/>
<w:LsdException Locked="false" Priority="46"
Name="List Table 1 Light Accent 4"/>
<w:LsdException Locked="false" Priority="47" Name="List Table 2 Accent 4"/>
<w:LsdException Locked="false" Priority="48" Name="List Table 3 Accent 4"/>
<w:LsdException Locked="false" Priority="49" Name="List Table 4 Accent 4"/>
<w:LsdException Locked="false" Priority="50" Name="List Table 5 Dark Accent 4"/>
<w:LsdException Locked="false" Priority="51"
Name="List Table 6 Colorful Accent 4"/>
<w:LsdException Locked="false" Priority="52"
Name="List Table 7 Colorful Accent 4"/>
<w:LsdException Locked="false" Priority="46"
Name="List Table 1 Light Accent 5"/>
<w:LsdException Locked="false" Priority="47" Name="List Table 2 Accent 5"/>
<w:LsdException Locked="false" Priority="48" Name="List Table 3 Accent 5"/>
<w:LsdException Locked="false" Priority="49" Name="List Table 4 Accent 5"/>
<w:LsdException Locked="false" Priority="50" Name="List Table 5 Dark Accent 5"/>
<w:LsdException Locked="false" Priority="51"
Name="List Table 6 Colorful Accent 5"/>
<w:LsdException Locked="false" Priority="52"
Name="List Table 7 Colorful Accent 5"/>
<w:LsdException Locked="false" Priority="46"
Name="List Table 1 Light Accent 6"/>
<w:LsdException Locked="false" Priority="47" Name="List Table 2 Accent 6"/>
<w:LsdException Locked="false" Priority="48" Name="List Table 3 Accent 6"/>
<w:LsdException Locked="false" Priority="49" Name="List Table 4 Accent 6"/>
<w:LsdException Locked="false" Priority="50" Name="List Table 5 Dark Accent 6"/>
<w:LsdException Locked="false" Priority="51"
Name="List Table 6 Colorful Accent 6"/>
<w:LsdException Locked="false" Priority="52"
Name="List Table 7 Colorful Accent 6"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="Mention"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="Smart Hyperlink"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="Hashtag"/>
<w:LsdException Locked="false" SemiHidden="true" UnhideWhenUsed="true"
Name="Unresolved Mention"/>
</w:LatentStyles>
</xml><![endif]--><!--[if gte mso 10]>
<style>
/* Style Definitions */
table.MsoNormalTable
{mso-style-name:"Table Normal";
mso-tstyle-rowband-size:0;
mso-tstyle-colband-size:0;
mso-style-noshow:yes;
mso-style-priority:99;
mso-style-parent:"";
mso-padding-alt:0in 5.4pt 0in 5.4pt;
mso-para-margin-top:0in;
mso-para-margin-right:0in;
mso-para-margin-bottom:8.0pt;
mso-para-margin-left:0in;
line-height:107%;
mso-pagination:widow-orphan;
font-size:11.0pt;
font-family:"Calibri",sans-serif;
mso-ascii-font-family:Calibri;
mso-ascii-theme-font:minor-latin;
mso-hansi-font-family:Calibri;
mso-hansi-theme-font:minor-latin;
mso-bidi-font-family:"Times New Roman";
mso-bidi-theme-font:minor-bidi;}
</style>
<![endif]-->
<br />
<div class="MsoNormal">
A while ago I created this image for thinking about how
machine learning systems tend to get deployed. In this figure, for <a href="http://ciml.info/dl/v0_99/ciml-v0_99-ch02.pdf">Chapter 2</a> of <a href="http://ciml.info/">CIML</a>, the left column shows a generic decision being
made, and the right column shows an example of this decision in the case of
advertising placement on a search engine we’ve built.</div>
<div class="MsoNormal">
<br /></div>
<div class="MsoNormal" style="text-align: center;">
<img alt="" height="352" src="data:image/png;base64,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" width="400" /></div>
<div class="MsoNormal">
<br /></div>
<div class="MsoNormal">
The purpose of the image at the time was basically thinking
of different types of “approximation” error, where we have some real world goal
(e.g., increase revenue) and design a machine learning system to achieve that.
The point here, which echoes a lot of the <a href="http://martin.zinkevich.org/rules_of_ml/rules_of_ml.pdf">Rules of Machine
Learning</a> by <a href="http://martin.zinkevich.org/">Martin Zinkevich</a> (who
knows much more about this than I do) writes about, is that it’s important to
recognize that there’s a whole lot of stuff that goes around any machine
learning system, and each piece puts an upper bound on what you can achieve.</div>
<div class="MsoNormal">
<br /></div>
<div class="MsoNormal">
A year or two ago, in talking to <a href="http://www.cs.utah.edu/~suresh/">Suresh Venkatasubramanian</a>, we
realized that it’s also perhaps an interesting way to think about different
places that “discrimination” might come into a system (I’ll avoid the term “bias”
because it’s overloaded here with the “bias/variance tradeoff”). By “discrimination”
I simply mean a situation in which some subpopulation is disadvantaged. Below
are some thoughts on how this might happen.</div>
<div class="MsoNormal">
<br /></div>
<div class="MsoNormal">
To make things more interesting (and navel-gaze-y), I’m
going to switch from the example of ad display to paper recommendations in a
hypothetical arxiv rss-feed-based paper recommendation system. To be absolutely
clear, this is very much a contrived, simplified thought example, and not meant
to be reflective of any decisions anyone or any company has made. (For the
purposes of this example, I will assume all papers on arxiv are in English.)</div>
<div class="MsoNormal">
<br /></div>
<div class="MsoNormal">
<br /></div>
<div class="MsoNormal">
<br /></div>
<span style="mso-bidi-font-family: Calibri; mso-bidi-theme-font: minor-latin;"><span style="mso-list: Ignore;">1.<span style="font: 7.0pt "Times New Roman";">
</span></span></span>We start with a <i style="mso-bidi-font-style: normal;">real-world goal</i>: helping people filter newly uploaded papers on arxiv.<br />
<div class="MsoListParagraphCxSpFirst" style="mso-list: l0 level1 lfo1; text-indent: -.25in;">
<br />
In stating this goal, we are explicitly making a value judgment of what
matters. In this case, one part of this value judgment is that it’s only <i style="mso-bidi-font-style: normal;">new papers</i> that are interesting,
potentially disadvantaging authors who have high quality older work. It also advantages
people who put their papers on arxiv, which is not a uniform slice of the
research population.<br />
<br style="mso-special-character: line-break;" /></div>
<span style="mso-bidi-font-family: Calibri; mso-bidi-theme-font: minor-latin;"><span style="mso-list: Ignore;">2.<span style="font: 7.0pt "Times New Roman";">
</span></span></span>We now need a <i style="mso-bidi-font-style: normal;">real-world mechanism</i> to achieve our goal: an iPhone app that shows
extracted information from a paper that users can thumbs-up or thumbs-down (or
swipe left/right as you prefer).<br />
<div class="MsoListParagraphCxSpMiddle" style="mso-list: l0 level1 lfo1; text-indent: -.25in;">
<br />
By deciding to build an iPhone app, we have privileged iPhone users over users
of other phones, which likely correlates both with country of residence and
economic status of the user. By designing the mechanism such that extracted
paper information is shown and a judgment is collected immediately, we are possibly
advantaging papers (and thereby the authors of those papers) whose
contributions can be judged quickly, or which seem flashy (click-bait-y). Similarly,
since human flash judgments may focus on less relevant features, we may be
biasing toward authors who are native English speakers, because things like
second language errors may disproportionately affect quick judgments.<br />
<br style="mso-special-character: line-break;" /></div>
<span style="mso-bidi-font-family: Calibri; mso-bidi-theme-font: minor-latin;"><span style="mso-list: Ignore;">3.<span style="font: 7.0pt "Times New Roman";">
</span></span></span>Next, set up a <i style="mso-bidi-font-style: normal;">learning problem</i>: online prediction of thumbs-up/down for papers
displayed to the user (essentially a contextual bandit learning problem).<br />
<div class="MsoListParagraphCxSpMiddle" style="mso-list: l0 level1 lfo1; text-indent: -.25in;">
<br />
I actually don’t have anything to say on this one. Open to ideas <span style="font-family: "segoe ui emoji" , sans-serif; mso-ascii-font-family: Calibri; mso-ascii-theme-font: minor-latin; mso-char-type: symbol-ext; mso-hansi-font-family: Calibri; mso-hansi-theme-font: minor-latin; mso-symbol-font-family: "Segoe UI Emoji";">😊</span>.<br />
<br style="mso-special-character: line-break;" /></div>
<span style="mso-bidi-font-family: Calibri; mso-bidi-theme-font: minor-latin;"><span style="mso-list: Ignore;">4.<span style="font: 7.0pt "Times New Roman";">
</span></span></span>Next, we define a mechanism for collecting data:
we will deploy a system and use epsilon-greedy exploration to collect data.<br />
<div class="MsoListParagraphCxSpMiddle" style="mso-list: l0 level1 lfo1; text-indent: -.25in;">
<br />
There are obviously repercussions to this decision, but I’m not sure any are discriminatory.
Had we chosen a more sophisticated exploration policy, this could possibly run
into discrimination issues because small populations might get “explored on”
more, <a href="https://arxiv.org/abs/1806.00543">potentially disadvantaging them</a>.<br />
<br style="mso-special-character: line-break;" /></div>
<span style="mso-bidi-font-family: Calibri; mso-bidi-theme-font: minor-latin;"><span style="mso-list: Ignore;">5.<span style="font: 7.0pt "Times New Roman";">
</span></span></span>From this data collection procedure, we decide
what data to log: paper title, authors, institution, abstract, and rating
(thumbs up/down).<br />
<div class="MsoListParagraphCxSpMiddle" style="mso-list: l0 level1 lfo1; text-indent: -.25in;">
<br style="mso-special-character: line-break;" /></div>
<div class="MsoListParagraphCxSpMiddle">
By choosing to record author and institution
(for instance), we are both opening up the possibility of discrimination against
certain authors or institutions, but, because many techniques for addressing discrimination
in machine learning <i style="mso-bidi-font-style: normal;">assume</i> that you
have access to some notion of protected category, we are also opening up the
possibility of remedying that. Similarly, by recording the abstract, we are (similar
but different to before) opening the possibility for discrimination by degree
of English proficiency.</div>
<div class="MsoListParagraphCxSpMiddle">
<br /></div>
<span style="mso-bidi-font-family: Calibri; mso-bidi-theme-font: minor-latin;"><span style="mso-list: Ignore;">6.<span style="font: 7.0pt "Times New Roman";">
</span></span></span>Given this logged data, we have to choose a data
representation: for this we’ll take text from title, authors, institution and
abstract, and then have features for the current user of the system (e.g., the
same features from their own papers), and a binary signal for thumbs up/down.<br />
<div class="MsoListParagraphCxSpMiddle" style="mso-list: l0 level1 lfo1; text-indent: -.25in;">
<br />
A major source of potential discrimination here comes from the features we use
of the current user. If the current user, for instance, only has a small number
of papers from which we can learn about the topics they care about, then the
system will plausibly work worse for them than for someone with lots of papers
(and therefore a more robust user profile).<br />
<br style="mso-special-character: line-break;" /></div>
<span style="mso-bidi-font-family: Calibri; mso-bidi-theme-font: minor-latin;"><span style="mso-list: Ignore;">7.<span style="font: 7.0pt "Times New Roman";">
</span></span></span>Next we choose a model family: for simplicity
and because we have a “cold start” problem, instead of going the full neural
network route, we’ll just use a bag-of-embeddings representation for the paper
being considered, and a bag-of-embeddings representation for the user, and
combine them with cross-product features into a linear model.<br />
<div class="MsoListParagraphCxSpMiddle" style="mso-list: l0 level1 lfo1; text-indent: -.25in;">
<br />
This is a fairly easy representation to understand. Because we’ve chosen a bag
of embeddings, this could plausibly underperform on topics/areas where the
keywords are separated by spaces (e.g., I heard a story once that someone who
works mostly on dependency parsing tends to get lots of papers suggested to them
to review by TPMS on decision tree models because of the overlap of the word “tree”).
It’s not clear to me that there are obvious discrimination issues here, but it
could be.<br />
<br style="mso-special-character: line-break;" /></div>
<span style="mso-bidi-font-family: Calibri; mso-bidi-theme-font: minor-latin;"><span style="mso-list: Ignore;">8.<span style="font: 7.0pt "Times New Roman";">
</span></span></span>Selecting the training data: in this case, the
training data is simply coming online, so the discussion in (3) applies.<br />
<div class="MsoListParagraphCxSpMiddle" style="mso-list: l0 level1 lfo1; text-indent: -.25in;">
<br style="mso-special-character: line-break;" /></div>
<span style="mso-bidi-font-family: Calibri; mso-bidi-theme-font: minor-latin;"><span style="mso-list: Ignore;">9.<span style="font: 7.0pt "Times New Roman";">
</span></span></span>We now train the model and select
hyperparameters: again, this is an online setting, so there’s really no
train/test split, so this question doesn’t really apply (though see the comment
about exploration in #4).<br />
<div class="MsoListParagraphCxSpMiddle" style="mso-list: l0 level1 lfo1; text-indent: -.25in;">
<br style="mso-special-character: line-break;" /></div>
<span style="mso-bidi-font-family: Calibri; mso-bidi-theme-font: minor-latin;"><span style="mso-list: Ignore;">10.<span style="font: 7.0pt "Times New Roman";">
</span></span></span>The model is then used to make predictions:
again, online, doesn’t really apply.<br />
<div class="MsoListParagraphCxSpMiddle" style="mso-list: l0 level1 lfo1; text-indent: -.25in;">
<br style="mso-special-character: line-break;" /></div>
<span style="mso-bidi-font-family: Calibri; mso-bidi-theme-font: minor-latin;"><span style="mso-list: Ignore;">11.<span style="font: 7.0pt "Times New Roman";">
</span></span></span>Lastly, we evaluate error: here, the natural is
0/1 loss on whether thumbs up/down was predicted correctly or not.<br />
<div class="MsoListParagraphCxSpLast" style="mso-list: l0 level1 lfo1; text-indent: -.25in;">
<br />
In choosing to evaluate our system based only on average 0/1 loss over the run,
we are potentially missing the opportunity to even <i style="mso-bidi-font-style: normal;">observe</i> systematic bias. An alternative would be to do things like
evaluating 0/1 error as a function of various confounding variables, like
institution prestige, author prolificity, some measure of nativism of the
language, etc. Similarly breaking the error down into features of the user for
similar reasons. Finally, considering not just error but separating out false
positive and false negatives can often reveal discriminatory structures not
otherwise obvious.</div>
<div class="MsoNormal">
<br /></div>
<div class="MsoNormal">
---------------------------- </div>
<div class="MsoNormal">
<br /></div>
<div class="MsoNormal">
I don’t think this analysis is perfect, and some things don’t
really apply, but I found it to be a useful thought exercise.</div>
<div class="MsoNormal">
<br /></div>
<div class="MsoNormal">
One thing very interesting about thinking about
discrimination in this setting is that there are two opportunities to mess up:
on the content provider (author) side and on the content consumer (reader) side.
This comes up in other places too: should your music/movie/etc. recommender
just recommend popular things to you or should it be fair to content providers
who are less well known? (Thanks to <a href="http://fernando.diaz.nyc/">Fernando
Diaz</a> for this example.)<br />
<br style="mso-special-character: line-break;" />
<br style="mso-special-character: line-break;" /></div>
halhttp://www.blogger.com/profile/02162908373916390369noreply@blogger.com2tag:blogger.com,1999:blog-19803222.post-33731915656276559872017-08-15T07:50:00.000-06:002017-08-15T09:47:13.154-06:00Column squishing for multiclass updatesScore-based multiclass classifiers typically have the following form: <i>x</i> is a d-dimensional input vector (perhaps engineered features, perhaps learned features), <i>A</i> is a d*k matrix, where k is the number of classes, and the prediction is given by computing a score vector <i>y=Ax.</i> (This is a blog post, not an arxiv paper, so I'm going to be a bit fast a loose with dimension ordering.) The predicted class is then taken as <i>argmax<sub>i</sub> y<sub>i</sub></i>.<br />
<br />
(Advanced thanks to my former Ph.D. student <a href="http://inductivebias.ml/">Abhishek Kumar</a> for help with this!) <br />
<br />
The question I wanted to answer, which I felt <i>must</i> have a known answer though I'd never seen it, is the following. Suppose (e.g., at training time), I know that the correct label is <i>i</i>, but the model is (perhaps) not predicting <i>i</i>. I'd like to change <i>A</i> as little as possible so that it predicts <i>i</i>, perhaps with an added margin of 1. If I measure "as little as possible" by l2 norm, and assume wlog that <i>i=1</i>, then I get:<br />
<br />
<i> min<sub>B</sub> ||B-A||<sup>2</sup> st (xB)<sub>1</sub>≥(xB)<sub>i</sub>+1 for all i ≥ 2</i>
<br />
<br />
This problem arises most specifically in Crammer et al., 2006, "<a href="http://jmlr.csail.mit.edu/papers/volume7/crammer06a/crammer06a.pdf">Online Passive-Aggressive Algorithms</a>", though appears elsewhere too.<br />
<br />
I'll show below (including python code using just numpy), a very efficient solution to this problem. (If this appears elsewhere and I simply was unable to find it, please let me know.)<br />
<br />
First, I'll make the following unproven assertion, though I'm pretty sure it'll go through (famous last words). The assertion is that any difference between <i>A</i> and <i>B</i> will be in the direction of <i>x</i>. In other words, the first row of <i>A</i> will likely move in the direction of <i>x</i> and the other rows of <i>A</i> will move away. Hand-wavy reason: because otherwise you increase the norm <i>||B-A||</i> without helping satisfy the constraints.<br />
<br />
In particular, I'll assume that <i>b<sub>i</sub>=a<sub>i</sub>+d<sub>i</sub>x</i>, where the dis are scalars.<br />
<br />
Given this, we can do a bit of algebra:<br />
<br />
<i> ||B-A||<sup>2</sup> = Σ<sub>i</sub> (b<sub>i</sub> - a<sub>i</sub>)<sup>2</sup> = Σ<sub>i</sub> (a<sub>i</sub> + d<sub>i</sub> x - a<sub>i</sub>)<sup>2</sup>= ||x||<sup>2</sup> Σ<sub>i</sub> d<sub>i</sub><sup>2</sup></i><br />
<br />
Since x is a constant, we really only care about minimizing the norm of the deltas.<br />
<br />
We can similarly rewrite the constraints to just say:<br />
<br />
<i>xb</i><i><i><sub>1</sub></i> ≥ xb</i><i><i><sub>i</sub></i> + 1 for all i ≥ 2</i><br />
<i>iff x(a</i><i><i><i><sub>1</sub></i></i> + d</i><i><i><i><sub>1</sub></i></i>x) ≥ x(a</i><i><i><sub>i</sub></i> + d</i><i><i><sub>i</sub></i>x) + 1 for all i ≥ 2</i><br />
<i>iff xa</i><i><i><i><sub>1</sub></i></i> + d</i><i><i><i><sub>1</sub></i></i> ||x||</i><i><i><sup>2</sup></i> ≥ xa</i><i><i><sub>i</sub></i> + d</i><i><i><sub>i</sub></i>||x</i><i><i>||<sup>2</sup></i> + 1 for all i ≥ 2</i><br />
<i><i>iff </i>d</i><i><i><i><sub>1</sub></i></i> ≥ d</i><i><i><sub>i</sub></i> + C</i><i><i><sub>i</sub></i> for all i ≥ 2</i><br />
<div class="separator" style="clear: both; text-align: center;">
<br /></div>
where <i>C</i><i><i><i><sub>i</sub></i></i> = [x(a</i><i><i><sub>i</sub></i>-a</i><i><i><i><sub>1</sub></i></i>)+1]/||x||</i><i><sup>2</sup></i> is independent of <i>d.</i><br />
<br />
<i></i>Now, we have a plausibly simpler optimization problem just over the <i>d</i> vector:<br />
<br />
<i> min<sub>d</sub> Σ<sub>i</sub> d<sub>i</sub><sup>2</sup> st d<sub>1</sub> ≥ d<sub>i</sub> + C<sub>i</sub> for all i ≥ 2</i><br />
<i><br /></i>
This was the place I got stuck. I felt like there would be some algorithm for solving this that involves sorting and projecting and whatever, but couldn't figure it out for a few days. I then asked current and former advisees, at which point <a href="http://inductivebias.ml/">Abhishek Kumar</a> came to my rescue :). He pointed me to the paper "<a href="http://pages.cs.wisc.edu/~brecht/papers/12.Bit.EtAl.HOTT.pdf">Factoring Non-negative Matrices with Linear Programs</a>" by Bittorf et al., 2012. It's maybe not obvious that this is all connect from the title, but they solve a very similar problem in Algorithm 5. All of the following is due to Abhishek:<br />
<br />
In particular, their Equation 11 has the form:<br />
<br />
<i> min<sub>x</sub> ||z-x||<sup>2</sup> st 0 ≤ x<sub>i</sub> ≤ x<sub>1</sub> for all i, x<sub>1</sub> ≤ 1</i><br />
<br />
My problem can be happed to this by a change of variables: <i>z=d+D</i>, where <i>D=[0, C</i><i><i><sub>2</sub></i>, C</i><i><i><sub>3</sub></i>, ..., C</i><i><i><sub>k</sub></i>]</i>. We also need to remove the lower and upper bounds. This means that their Algorithm 5 can be used to solve my problem, but with all of the [0,1] projection steps removed. For completeness, here is their algorithm:<br />
<br />
<div class="separator" style="clear: both; text-align: center;">
<a href="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEgn-PwLKb0RNUNjZpEXs9dMniuomIpNQxAu9YRm_64GCFCg_oqAJ4vcGpXh0OE6yCEbJ7pqvPdzLhZwgPkf9i4qIeRnH0LMermUQ90TG8tZ7xQQoxszXOOxMM74LmkWc2XNnaCAag/s1600/column_squishing.png" imageanchor="1" style="margin-left: 1em; margin-right: 1em;"><img border="0" data-original-height="598" data-original-width="1161" height="205" src="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEgn-PwLKb0RNUNjZpEXs9dMniuomIpNQxAu9YRm_64GCFCg_oqAJ4vcGpXh0OE6yCEbJ7pqvPdzLhZwgPkf9i4qIeRnH0LMermUQ90TG8tZ7xQQoxszXOOxMM74LmkWc2XNnaCAag/s400/column_squishing.png" width="400" /></a></div>
<br />
<br />
Putting this all together, we arrive at some python code in <a href="http://hal3.name/column_squishing.py">column_squishing.py</a> for solving my multiclass problem. Here's an example of running it:<br />
<br />
<br />
<pre>≫ A = np.random.randn(3,5)
≫ x = np.random.randn(5)
≫ A.dot(x)
array([ 0.90085352, 2.25573249, 0.25974194])
</pre>
<br />
So currently label "1" is winning by a big margin. Let's make each label win by a margin of one, one at a time: <br />
<br />
<pre>≫ multiclass_update(A, x, 0).dot(x)
array([ 2.078293 , 1.078293 , 0.25974194])</pre>
<pre> </pre>
<pre>≫ multiclass_update(A, x, 1).dot(x) </pre>
<pre>array([ 0.90085352, 2.25573249, 0.25974194])</pre>
<pre> </pre>
<pre>≫ multiclass_update(A, x, 2).dot(x) </pre>
<pre>array([ 0.80544265, 0.80544265, 1.80544265])
</pre>
<pre> </pre>
Hopefully you find this helpful. If you end up using it, please make some sort of acknowledgement to this blog post and definitely please credit Abhishek.halhttp://www.blogger.com/profile/02162908373916390369noreply@blogger.com5tag:blogger.com,1999:blog-19803222.post-18114192719057487332017-04-12T08:07:00.000-06:002017-04-12T08:58:32.820-06:00Humans can still extort more money from me than machines canLike lots of folks, I wonder sometimes about AI and jobs. I'm neither a believer that there's a catastrophe coming up, nor am I a believer that everything will magically work out and we're not entering a world with new forms of inequities.<br />
<br />
I did have an experience recently that made me think somewhat differently about what sorts of jobs are at risk.<br />
<br />
I was visiting family in LA, and renting a car (because LA). I can't remember what company it was, but they didn't have "live people" at the booth. Instead they had a row of kiosks, each with a monitor and camera and old school phone that---I kid you not---looked like:<br />
<div class="separator" style="clear: both; text-align: center;">
<a href="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEjnK8KzbhXq3uw9F-e9Mtp7sHjGAD-KcgOaoK9esRRqvUt3wlchMv3mb1nh8jIpoJ-zJ07PKLj6hnDhk2534V8iQhpjHKDrRW1zuHo225o2qxCXX-WG7ME0GGLUqnB5ItI5scexnA/s1600/s0020949_sc7.jpeg" imageanchor="1" style="margin-left: 1em; margin-right: 1em;"><img border="0" src="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEjnK8KzbhXq3uw9F-e9Mtp7sHjGAD-KcgOaoK9esRRqvUt3wlchMv3mb1nh8jIpoJ-zJ07PKLj6hnDhk2534V8iQhpjHKDrRW1zuHo225o2qxCXX-WG7ME0GGLUqnB5ItI5scexnA/s1600/s0020949_sc7.jpeg" /></a></div>
Complete with curly cord.<br />
<br />
So how does this work? You go up to the device, and it auto-dials a real person. This person lives who knows were, though in my case he had a nice painted backdrop of some nature scene.<br />
<br />
They ask about your reservation, look it up, etc., and while looking it up, he asks what I'm doing in LA. Oh I'm visiting my mom. Blah blah blah.<br />
<br />
I then have to hold my drivers license up to the camera, we chat some more about how LA traffic is horrible. Blah blah blah.<br />
<br />
Throughout this whole conversation, I'm thinking: this would be so much easier if this thing were automated. I don't mean automated like "chatbot", I mean automated like the kiosks at airports, where I just push a bunch of buttons and then get a ticket (or in this case, car).<br />
<br />
And then, as you're all expecting, he asks me if I want to buy insurance.<br />
<br />
No machine would ever under any circumstances be able to sell me insurance. It would show a screen, ask me to click if I wanted it or not, and I would immediately press the "No" button.<br />
<br />
Now, I'm kind of a pushover, and so, especially since this guy was nice, had asked about my mom, shared complaints about traffic together, etc., when he asked me if I wanted insurance, it was <i>much harder</i> to say no. I did say no, though I'm pretty sure that me-five-years-ago might not have.<br />
<br />
Whatever company this is, I'm sure they did a study. They looked at how much they'd save by having automated systems rather that some folks in presumably a part of the country/world with much lower cost of living than LA. And the main trade-off was almost certainly that a machine isn't going to be nearly as successful at upselling insurance or car model or whatever as a person. And clearly at the end of the day, they decided that automation wasn't worth it here.<br />
<br />
(Insert all sorts of analogies to other jobs/areas of life.)<br />
<br />
EDIT 11am Eastern 12 Apr 2018: It occurred to me after I hit "post" that it's worth highlighting a major implicit caveat: the fact that the alleged study showed that it was worth putting money into human interaction is probably largely due to whatever the car rental company knew about, for instance, the economic status of their average customer. Where this needle falls---for instance if no few customers will choose to or can afford to be upsold---will have a significant impact on what the results of such a study will be, and will therefore vary across industries.<br />
<br />
<br />halhttp://www.blogger.com/profile/02162908373916390369noreply@blogger.com1tag:blogger.com,1999:blog-19803222.post-51757482628976676162017-04-03T06:49:00.000-06:002017-04-03T06:49:58.365-06:00Structured prediction is *not* RL<div class="separator" style="clear: both; text-align: center;">
</div>
<div style="margin-left: 1em; margin-right: 1em;">
</div>
It's really strange to look back now over the past ten to fifteen years and see a very small pendulum that no one really cares about swing around. I've spent the last ten years trying to convince you that structured prediction is RL; now I'm going to tell you that was a lie :).<br />
<br />
<b>Short Personal History</b> <br />
<br />
Back in 2005, John Langford, Daniel Marcu and I had a workshop paper at NIPS on <a href="http://hal3.name/docs/daume05search.pdf">relating structured prediction to reinforcement learning</a>. Basically the message of that paper is: you can cast structured prediction as RL, and then use off-the-shelf RL techniques like <a href="https://www.cs.cmu.edu/~jcl/presentation/RL/RL.ps">conservative policy iteration</a> to solve it, and this works pretty well. This view arose, for me, main from Collins and Roark's <a href="http://www.aclweb.org/anthology/P04-1015">incremental structured perceptron</a> model, but I'm sure it dates back much further than that (almost certainly there's work from neural nets land in the 80s; I'd certainly appreciate pointers!). This led <a href="https://twitter.com/haldaume3/status/832018253731880960">eventually</a> to <a href="https://arxiv.org/pdf/0907.0786">Searn</a>, and then in the early 2010s, I went around a bunch of places (ACL, AAAI, ICML, etc.), <a href="https://pdfs.semanticscholar.org/2389/9124a85eea50fd955812a33c50a663ece559.pdf">espousing</a> <a href="https://pdfs.semanticscholar.org/5d75/912569b91068b2e08b9ea036136d4d55765e.pdf">connections</a> between structured prediction and (inverse) reinforcement learning (<a href="http://hal3.name/11-08-spirl.pdf">slides</a>).<br />
<br />
This kinda sorta caught on in a very small subcommunity.<br />
<br />
One thing I should have realized looking back ten years is that you should not try to sell a hammer to hammer manufacturers, but rather to people with weird nails. I spent too much time and energy trying to convince the "traditional structured prediction" crowd (the CRF and M3N and SVM<sup>struct</sup> folks) that the "RL" view of structured prediction was awesome. That was a losing strategy, unfortunately. (Although I learned a few nights ago at dinner that this might be changing now!)<br />
<br />
But now everything has changed. To some degree, Ross, Gordon and Bagnell's <a href="https://www.cs.cmu.edu/~sross1/publications/Ross-AIStats11-NoRegret.pdf">DAgger</a> is a successful, better successor to Searn, and for years I had gone around telling everyone DAgger is my favorite algorithm ever (it consistently outperforms Searn's in their---and my---experiments, and is really really easy to implement and has stronger-ish theory). And then DAgger (or more precisely <a href="https://www.ri.cmu.edu/pub_files/2015/1/Venkatraman.pdf">DAD</a>) gets <a href="https://nlpers.blogspot.com/2016/03/a-dagger-by-any-other-name-scheduled.html">renamed/rebranded</a> as "<a href="http://papers.nips.cc/paper/5956-scheduled-sampling-for-sequence-prediction-with-recurrent-neural-networks.pdf">scheduled sampling</a>" (though you should read Marc'Aurelio's comment, which is very on point), and now these ideas are everywhere, particularly in sequence-to-sequence neural transduction models.<br />
<br />
<b>Nowadays</b> <br />
<br />
In the past year or two, there's been a <a href="https://scholar.google.com/scholar?as_ylo=2016&q=reinforcement+learning+seq2seq&hl=en&as_sdt=0,39">flurry of work</a> applying not just imitation learning algorithms like DAgger to neural models for structured prediction problems, but also just applying straight-up RL algorithms (like reinforce, policy gradient, or actor/critic) to them. The important point is that while people have tried to do things like neural CRFs, etc., the basic sequence-to-sequence style model naturally fits a search-based structured prediction (aka RL-ish) view.<br />
<br />
But these tasks are <i>not</i> the same and, in fact, structured prediction is <i>much</i> simpler, and I think we need to develop algorithms that take that into account.<br />
<br />
The biggest difference is that in (all or at least almost all) structured prediction problems, conditioned on the input <i>x</i>, the world is <i>known</i>, <i>deterministic</i> and therefore reversible and/or fully-explorable (modulo limits of computation)<i>.</i> This is generally not true in RL, and one of biggest challenges in RL is that once you take an action, you cannot un-take that action, and you cannot try out other alternatives.<br />
<br />
That is to say: computation aside, in structured prediction, conditioned on <i>x</i>, you can build out the <i>entire search tree</i> and do whatever the heck you want with it. (Of course "computation aside" makes no sense in a SP setting because the whole difficulty of SP is computation.) In fact, one of the big advantages to things like CRFs is effectively that they <i>do</i> build out the entire search tree, at least implicitly, which is possibly precisely because of the limited expressivity of features.<br />
<br />
This observation is probably perfectly obvious to most NLP folks. In a sense, the <a href="https://transacl.org/ojs/index.php/tacl/article/viewFile/27/4">semantic</a> <a href="http://www.aclweb.org/website/old_anthology/D/D13/D13-1160.pdf">parsing</a> <a href="http://www.jayantkrish.com/papers/jayantk-emnlp2012.pdf">crowd</a> has been doing something reinforcement-learning like for quite some time. You produce a semantic parse, run it against a database, check if you get the correct answer or not. If so, positive reward; if not, negative. But no one (as far as I know) just produces one parse: you produce a beam of a bunch of parses and try them all. This is definitely <i>not</i> something you can do in standard RL, but it <i>is</i> something you can do in structured prediction.<br />
<br />
In my mind, this was (and continues to be) one of the major weakness of the Searn/DAgger approach to structured prediction that continues to be a problem in applying standard RL algorithms to structured prediction. In a real sense, I think this is something that <a href="http://www.aclweb.org/anthology/P04-1015">incremental perceptron</a>, broken <a href="http://hal3.name/docs/daume05laso.pdf">LaSO</a> and not-broken <a href="http://web.engr.oregonstate.edu/~afern/papers/beam-icml07.pdf">LaSO-BST</a>, and <a href="https://arxiv.org/pdf/1606.02960">seq2seqLaSO</a> got right that the more RL-ish approaches got wrong. (This continues to be true in bandit structured prediction, which will get a separate post in the maybe-near future.) One approach that blends the two to some degree is <a href="https://timvieira.github.io/doc/2017-tacl-pruning.pdf">Vieira and Eisner's recent paper</a> that uses dynamic programming and change propogation within LOLS (which is effectively a variant of <a href="https://arxiv.org/abs/1406.5979">AggreVaTe</a>, a follow-on to DAgger) to learn to prune (it's not obvious to me how to generalize this to other tasks yet).<br />
<br />
<b>Why the Gap?</b><br />
I don't think this gap is an accident, and I think there are essentially two reasons it exists.<br />
<br />
First, as suggested above, is the question of computation. If you're willing to say "I don't care about computation" then you might as well put yourself in CRF world where life is (relatively) easy, at least if you want to do analysis. If you do care about computation, then doing the "purely greedy" thing is very natural and then you can say "well I know I'm computationally efficient because I'm greedy, and now I can focus entirely on the statistical problem." Once you're willing to spend a bit more computation, you have to figure out how to "charge" yourself properly for that computation. That is, you enter a world where there's a trade-off between statistics and computation (though not in the usual sense) and it's not at all clear how to balance that. It's also hard to convince myself that it's better to spend ten times longer on a single structured example than to do ten different examples. This is a question I've been interested in since <a href="http://hal3.name/docs/daume06thesis.pdf">my dissertation (p44)</a> but have made basically zero progress on:<br />
<br />
<div class="separator" style="clear: both; text-align: center;">
<a href="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEjyVcO8Jfxrm0tdEFYz7D3owZv3a-kPJRoY424zmFThZZfn3cL8wOFrSzSjTVbMsClPLoIKh6kAZ585z7ZhbzcEc1c8hO2KHXFMYHgjlzV6FmioOXEjqOHi3lt7BXv_1JVfvr3xzw/s1600/thesis.png" imageanchor="1" style="margin-left: 1em; margin-right: 1em;"><img border="0" height="258" src="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEjyVcO8Jfxrm0tdEFYz7D3owZv3a-kPJRoY424zmFThZZfn3cL8wOFrSzSjTVbMsClPLoIKh6kAZ585z7ZhbzcEc1c8hO2KHXFMYHgjlzV6FmioOXEjqOHi3lt7BXv_1JVfvr3xzw/s400/thesis.png" width="400" /></a></div>
(Note that I don't currently agree with the entirety of this passage---in particular, the complexity argument is somewhat broken---but I think the basic idea is right.)<br />
<br />
<div class="separator" style="clear: both; text-align: center;">
</div>
Second, I think there's a bit of a looking-under-the-lamppost effect that's not easy to ignore. Here, the lamppost is mainly a computational efficiency lamppost, and secondarily a convenience lamppost. Greedy search is really fast. Even compared to beam search with a beam size of K, greedy is often much more than K times faster because you don't have bookkeeping overhead. And it's <i>way</i> easier to implement greedy solutions than non-greedy, especially in neural land if you want things to be efficient on a GPU. And often toolkits have greedy already implemented for you. This obviously isn't un-recoverable, but runs into the problem of: if I can do 50 sentences greedily on my GPU in the time it would take to do beam-10 search for one of the sentences, is the computation really going to come off in my favor?<br />
<br />
As always, I'd appreciate pointers to work that I don't know about that addresses any of these challenges!<br />
<br />halhttp://www.blogger.com/profile/02162908373916390369noreply@blogger.com6tag:blogger.com,1999:blog-19803222.post-89968821886122303162017-03-27T11:35:00.002-06:002017-03-27T11:35:53.733-06:00Initial thoughts on fairness in paper recommendation?There are a handful of definitions of "fairness" lying around, of which the most common is disparate impact: the rate at which you hire members of a protected category should be at least 80% of the rate you hire members not of that category. (Where "hire" is, for our purposes, a prediction problem, and 80% is arbitrary.) DI has all sorts of issues, as do many other notions of fairness, but all the ones I've seen rely on a pre-ordained notion of "protected category".<br />
<br />
I've been thinking a lot about something many NLP/ML people have thought about in their musing/navel-gazing hours: something like a recommender system for papers. In fact, <a href="https://cs.stanford.edu/~pliang/">Percy Liang</a> and I build something like this a few years ago (called Braque), but it's now defunct, and its job wasn't really to recommend, but rather to do offline search. Recommendation was always lower down the TODO list. I know others have thought about this a lot because over the last 10 years I've seen a handful of proposals and postdoc ads go out on this topic, though I don't really know of any solutions.<br />
<br />
A key property that such a "paper recommendation system" should have is that it be fair.<br />
<br />
But what does fair mean in this context, where the notion of "protected category" is at best unclear and at worst a bad idea? And to <i>whom</i> should it be fair?<br />
<br />
Below are some thoughts, but they are by no means complete and not even necessarily good :P.<br />
<br />
In order to talk about fairness of predictions, we have to first define what is being predicted. To make things concrete, I'll go with the following: the prediction is whether the user wants to read the entire paper or not. For instance, a user might be presented with a summary or the abstract of the paper, and the "ground truth" decision is whether they choose to read the rest of the paper.<br />
<br />
The most obvious fairness concept is <i>authorship fairness:</i> that whether a paper is recommended or not should be independent of who the authors are (and what institutions they're from). On the bright side, a rule like this attempts to break the rich-get-richer effect, and means that even non-famous authors' papers get seen. On the dark side, authorship is actually a useful feature for determining how much I (as a reader) trust a result. Realistically, though, no recommender system is going to model whether a result is trustworthy: just that someone finds a paper interesting enough to read beyond the abstract. (Though the two are correlated.)<br />
<br />
A second obvious but difficult notion of fairness is that performance of the recommender system should not be a function of, eg., how "in domain" the paper is. For example, if our recommender system relies on generating parse trees (I know, comical, but suppose...), and parsing works way better on NLP papers than ML papers, this shouldn't yield markedly worse recommendations for ML papers. Or similarly, if the underlying NLP fares worse on English prose that is slightly non-standard, or slightly non-native (for whatever you choose to be "native"), this should not systematically bias against such papers.<br />
<br />
A third notion of fairness might have to do with underlying popularity of topics. I'm not sure how to formalize this, but suppose there are two topics that anyone ever writes papers about: deep learning and discourse. There are far more DL papers that discourse papers, but a notion of fairness might establish that they be recommended at similar rates.<br />
<br />
This strong rule seems somewhat dubious to me: if there are lots of papers on DL then probably there are lots of readers, and so probably DL papers should be recommended more. (Of course it could be that there exists an area where tons of papers get written and none get read, in which case this wouldn't be true.)<br />
<br />
A weaker version of this rule might state conditions of one-sided error rates. Suppose that every time d discourse paper is recommended, it is read (high precision), but that only about half of the recommended DL papers get read (low precision). Such a situation might be considered unfair to discourse papers because tons of DL papers get recommended when they shouldn't, but not so for discourse papers.<br />
<br />
Now, one might argue that this is going to be handled by just maximizing accuracy (aka click-through rate), but this is not the case if the number of people who are interested in discourse is dwarfed by the number interested in DL. Unless otherwise constrained, a system might completely forgo performance on those interested in discourse in favor of those interested in DL.<br />
<br />
This is all fine, except that the world doesn't consist of just DL papers and just discourse papers (and nary a paper in the intersection, sorry <a href="http://www.cc.gatech.edu/grads/y/yji37/papers/ji-acl-2014.pdf">Yi and Jabob</a> :P). So what can we do then?<br />
<br />
Perhaps a strategy is to say: I should <i>not</i> be able to predict the accuracy of recommendation on a specific paper, given its contents. That is: just because I know that a paper includes the words "discourse" and "RST" shouldn't tell me anything about what the error rate is on this paper. (Of course it <i>does</i> tell me something about the recommendations I would make on this paper.) You'd probably need to soften this with some empirical confidence intervals to handle the fact that many papers will have very few observations. You could also think about making a requirement/goal like this simultaneously on both false positives and false negatives.<br />
<br />
A related issue is that of bubbles. I've many times been told that one of my (pre-neural-net) papers was done in neural nets land ten years ago; I've many times told-or-wanted-to-tell the opposite. Both of these are failures of exploration. Not out of malice, but just out of lack-of-time. If a user chooses to read papers if and only if they're on DL, should a system continue to recommend non-DL papers to them? If so, why? This directly contradicts the notion of optimizing for accuracy.<br />
<br />
Overall, I'm not super convinced by any of these thoughts enough to even try to really formalize them. Some relevant links I found on this topic:<br />
<ul>
<li><a href="http://coen.boisestate.edu/piret/projects/recsys-bias/">Bias in Recommender Systems</a></li>
<li><a href="http://ceur-ws.org/Vol-1247/recsys14_poster10.pdf"><span id="goog_196494750"></span>Correcting Popularity Bias by Enhancing Recommendation Neutrality<span id="goog_196494751"></span></a> </li>
</ul>
halhttp://www.blogger.com/profile/02162908373916390369noreply@blogger.com2tag:blogger.com,1999:blog-19803222.post-70338785029637238842017-03-16T08:07:00.001-06:002017-03-16T08:40:28.051-06:00Trying to Learn How to be Helpful (IWD++)Over the past week, in honor women of this <a href="https://www.internationalwomensday.com/">International Women's Day</a>, I had a several posts, broadly around the topic of women in STEM. Previous posts in this series include: <a href="https://nlpers.blogspot.com/2017/03/awesome-people-bonnie-dorr-iwd.html">Awesome People: Bonnie Dorr</a>, <a href="https://nlpers.blogspot.com/2017/03/awesome-people-ellen-riloff-iwd.html">Awesome People: Ellen Riloff</a>, <a href="https://nlpers.blogspot.com/2017/03/awesome-people-lise-getoor-iwd.html">Awesome People: Lise Getoor</a>, <a href="https://nlpers.blogspot.com/2017/03/awesome-women-karen-sparck-jones-iwd.html">Awesome People: Karen Spärck Jones</a> and <a href="https://nlpers.blogspot.com/2017/03/awesome-people-kathy-mckeown-iwd.html">Awesome People: Kathy McKeown</a>. (Today's is delayed one day, sorry!)<br />
<br />
I've been incredibly fortunate to have a <i>huge</i> number of influential women in my life and my career. Probably the other person (of any gender) who contributed to my career as much as those above is my advisor, <a href="https://www.isi.edu/~marcu/">Daniel</a>. I've been amazingly supported by these amazing women, and I've been learning and thinking and trying to do a lot over the past few years to do what I can to support women in our field.<br />
There are a lot of really good articles out there on how to be a male ally in "tech." Some of these are more applicable to academia than others and I've linked to a few below.<br />
<br />
Sometimes "tech" is not the same as "academia", and in the context of the academy, easily the best resource I've seen is Margaret Mitchell's writeup on a <a href="http://m-mitchell.com/gender/papers/5Aug2016.pdf">Short Essay on Retaining and Increasing Gender Diversity, with Focus on the Role that Men May Play</a>. You should go read this now.<br />
<br />
No really, go read it. I'll be here when you get back.<br />
<br />
On Monday I attended a <a href="https://local.anitaborg.org/event/role-male-allies-advocates-mentors-retaining-women-tech/">Male Allies in Tech workshop</a> put on by <a href="http://anitaborg.org/">ABI</a>, with awesome organization by <a href="https://twitter.com/rosariorobinson">Rose Robinson</a> and <a href="https://twitter.com/laurennor">Lauren Murphy</a>, in which many similar points were made. This post is basically a summary of the first half of that workshop, with my own personal attempt to try to interpret some of the material into an academic setting. Many thanks to especially to <a href="https://twitter.com/NataliasTweeet">Natalia Rodriguez</a>, <a href="https://twitter.com/therealeringrau">Erin Grau</a>, <a href="https://twitter.com/RFagiri">Reham Fagiri</a> and <a href="https://twitter.com/tastelifewrite">Venessa Pestritto</a> on the Women perspectives panel, and Dan Storms, Evin Robinson, Chaim Haas and Kip Zahn on the Men perspective panel (especially <a href="https://twitter.com/EvinRobinson">Evin Robinson</a>!).<br />
<br />
The following summary of the panels has redundancy with many of Margaret's points, which I have not suppressed and have tried to highlight.<br />
<ol>
<li style="margin-bottom: 1em;"><b>Know that you're going to mess up and own it.</b> I put this one first because I'm entirely sure that even in the writing this post, I'm going to mess up. I'm truly uncomfortable writing this (and my fingernails have paid the price) because it's not about me, and I really don't want to center myself. On the other hand, I also think it's important to discuss how men (ie me) can try to be helpful, and shying away from discussion also feels like a problem. The only place I feel like I can honestly speak from is my own experience. This might be the worst idea ever, and if it is, I hope someone will tell me and talk to me about it. So please, feel free to email me, message me, come find me in person or whatever. <br /> </li>
<li style="margin-bottom: 1em;"> Pretty much the most common thing I've heard, read, whatever, is: <b>listen to and trust</b> <b>women</b>. Pretty much all the panelists at the workshop mentioned in this in some form, and Margaret mentions this in several places. As an academic, though, there's more that I've tried to do: read some papers. There's lots of research on the topic of things like unconscious bias in all sorts of settings, and studies of differences in how men and women are cited, and suggested for talks, and everything else under the sun. A reasonable "newbie" place to start might be Lean In (by Sheryl Sandberg) which, for all the issues that it has, provides some overview of research and includes citations to literature. But in general, doing research and reading papers is something I know how to do, so I've been trying to do it. Beyond being important, I've honestly found it really intellectually engaging.
</li>
<li style="margin-bottom: 1em;"> Another very frequently raised topic on the panels, and something that Margeret mentions too, is to <b>say something when you see or hear something sexist</b>. Personally, I'm pretty bad at thinking of good responses in these cases: I'm a very non-type-A person, I'm not good with confrontation, and my brain totally goes into "flight" mode. I've found it really useful to have a cache of go-to responses. An easy one is something like "whoah" or just "not cool" whose primary benefit is being easy [3]. Both seem to work pretty well, and take very little thought/planning. A more elaborate alternative is to ask for clarification. If someone says something sexist, ask what's meant by that. Often in the process of trying to explain it, the issue becomes obvious. (I've been on the receiving side of both such tactics, too, and have found them both effective there as well.)<br /><br />Another standard thing in meetings is for men to restate what a woman has stated as their own idea. A suggested response from <a href="https://www.linkedin.com/in/rosariorobinson">Rose Robinson</a> (one of the organizers) at the workshop is "I'm so glad you brought that up because maybe it wasn't clear when [woman] brought it up earlier." I haven't tried this yet, but it's going into my collection of go-to responses so I don't have to think too much. I'd love to hear other suggestions!
</li>
<li style="margin-bottom: 1em;"> A really interesting suggestion from the panel at the workshop was "go find a woman in your organization with the same position as you and tell her your salary." That said, I've heard personally from two women at two different universities that they were told they could not be given more of a raise because then they'd be making more than (some white guy). I'm not sure what I can do about cases like that. A related topic is startup: startup packages in a university are typically <i>not</i> public, so a variant of this is to tell your peers what your startup was.
</li>
<li style="margin-bottom: 1em;"> There were a lot of suggestions around the idea of making sure that your company's content has <b>broad representation</b>; I think in academic this is closely related to the first three of Margaret's points about suggesting women for panels, talks or interviews in your stead. I would add leadership roles to that list. One thing I've been trying to do when I'm invited to regular seminar series is to look at their past speakers and decide whether I would be contributing to the problem by accepting. This is harder for one-off things like conference talks/panels (because there's often no history), but even in those cases it's easy enough to ask if I'll be on an all-male panel. In cases where I've done this, the response has been positive. I've also been trying to be more openly intentional recently: if I do accept something, I'll try to explicitly say that I'm accepting because I noticed that past speakers were balanced. Positive feedback is good. A personally useful thing I did was write template emails for <a href="http://hal3.name/tmp/response_emails.txt">turning down invitations or asking for more information</a>, with a list of researchers from historically excluded groups in CS (including but not limited to women) who could be invited in my stead. I almost never send these exactly as is, but they give me a starting point.<br /><br />There's a dilemma here: if every talk series, panel, etc., were gender balanced, women would be spending all their time going around giving talks and would have less time for research. I don't have a great solution here. (I do know that a non-solution is to be paternalistic and make decisions for other people.) One option would be to pay honoraria to women speakers and let the "market" work. This doesn't address the dilemma fully (time != money), but I haven't heard of or found other ideas. Please help!<br /><br />
Turning down invitations to things as an academic is really hard. I recognize my relative privilege here that I already have tenure and so the cost to me for turning down this or that is pretty low in comparison to someone who is still a Ph.D. student or an untenured faculty member. That is to say: it's easy for me to say that I'm willing to take a short term negative reward (not giving a talk) in exchange for a long term very positive reward (being part of a more diverse community that both does better science and is also more supportive and inclusive). If I were still pre-tenure, this would definitely get clouded with the problem that it's great if there's a better environment in the future but not so great for me if I'm not part of it. On the other hand, pre-tenure is definitely a major part of the leaky pipeline, and so it's also really important to try to be equitable here. Each person is going to have to find a balance that they're comfortable with.<br /><br />
One last thought on this topic is something that I was very recently inspired to think about by Hanna Wallach. My understanding is that she, like most people, cannot accept honoraria as part of a company, and so she recently started asking places to donate her honoraria to good causes. I <i>can</i> accept honoraria for talks, which hurts the pipeline, but perhaps by donating these funds to organizations like ABI or BlackGirlsCode, I can try to help other parts of the pipeline. (There are tons of organizations out there I've thought about supporting; I like BGC for original intersectionality reasons.)
</li>
<li style="margin-bottom: 1em;"> I've been working hard to <b>follow women on social media</b> (and to follow members of other historically excluded groups, including women). This has been super valuable to me for expanding my views of tons of topics.
</li>
<li style="margin-bottom: 1em;"> The final topic at the workshop was a talk by the two authors of a new book on <b>how and why men can mentor women</b> called <a href="http://www.publishersweekly.com/978-1-62956-151-6">Athena Rising</a>. This was really awesome. Mentoring in tech is different than advising in academia, but not <i>that</i> different. Or at least there are certainly some parallels. Looking back at Hal-a-few-years ago, I very much had fallen into the trap of "okay I advise a diverse group of PhD students ergo I'm supporting diversity." This is painfully obvious now when I re-read old grant proposals. A consistent thing I've heard is that this is a pretty low bar, especially because women who do the extra required to get to our PhD program are really really amazing.<br /><br />I still think this is an important factor, but this discussion at the workshop made me realize that I can also go out and learn how to be a better advisor, especially to students whose live experiences are very different than my own. And that it's okay if students don't want the same path in life that I do: "hone don't clone" was the catch-phrase here. This discussion reminded me a comment one of the PhD students made to me after going to Grace Hopper: she really appreciated it because she could ask questions there that she couldn't ask me. I think there will always be such questions (because my lived experience is different), but I've decided to try to close the gap a bit by learning more here.
</li>
<li style="margin-bottom: 1em;"> Finally (and really, thank you if you've read this far), a major problem that was made apparent to me by Bonnie Webber is that one reason that women receive fewer awards in general is because <b>women are nominated for fewer awards</b> (note: this is not the <i>only</i> reason). Nominating women for awards is a <i>super easy</i> thing for me to do. It costs a few hours of my time to nominate someone for an award, or to write a letter (of course for serious awards, it's far more than a few hours to write a letter, but whatev). This includes internal awards at UMD, as well as external awards like ACL (or ACM or whatever) fellows, etc. Whenever I get an email for things like this, I'm trying to think about: who could I nominate for this that might otherwise be overlooked (Margaret's point on page 2!).
</li>
</ol>
I promised some other intro resources; I would suggest:
<br />
<ol>
<li> <a href="http://geekfeminism.wikia.com/wiki/Allies">GeekFeminism: Allies</a>
</li>
<li> <a href="http://geekfeminism.wikia.com/wiki/Resources_for_allies">GeekFeminism: Resources for Allies</a>
</li>
<li> <a href="http://geekfeminism.wikia.com/wiki/Good_sexism_comebacks">GeekFeminism: Good sexism comebacks</a>
</li>
<li> <a href="http://everydayfeminism.com/2015/01/male-feminist-rules-to-follow/"> Everyday Feminism: Male Feminist Rules to Follow</a>
</li>
<li> <a href="http://geekfeminism.wikia.com/wiki/Allies_workshop">GeekFeminism: Allies Workshop</a>
</li>
</ol>
Like I said at the beginning, what I really hope is that people will reply here with (a) suggestions for things they've been trying that seem to be working (or not!), (b) critical feedback that something here is really a bad idea and that something else is likely to be much more effective, (c) and general discussion about the broad issues of diversity and inclusion in our communities.<br />
<br />
Because of the topic of the workshop, this is obviously focused in particular on women, but the broader discussion needs to include topics related to <i>all</i> historically excluded groups because what works for one does not necessary work for another. Especially when intersectionality is involved. Rose Robinson ended the ABI Workshop saying "To get to the same place, women have to do extra. And Black women have to do extra extra." What I'm trying to figure out is what extra <i>I</i> can do to try to balance a bit more. So please, please, help me!halhttp://www.blogger.com/profile/02162908373916390369noreply@blogger.com12tag:blogger.com,1999:blog-19803222.post-2785187211117200892017-03-14T07:41:00.000-06:002017-03-14T07:41:17.655-06:00Awesome people: Kathy McKeown (IWD++)To honor women this <a href="https://www.internationalwomensday.com/">International Women's Day</a>, I have a several posts, broadly around the topic of women in STEM. Previous posts in this series include: <a href="https://nlpers.blogspot.com/2017/03/awesome-people-bonnie-dorr-iwd.html">Awesome People: Bonnie Dorr</a>, <a href="https://nlpers.blogspot.com/2017/03/awesome-people-ellen-riloff-iwd.html">Awesome People: Ellen Riloff</a>, <a href="https://nlpers.blogspot.com/2017/03/awesome-people-lise-getoor-iwd.html">Awesome People: Lise Getoor</a> and <a href="https://nlpers.blogspot.com/2017/03/awesome-women-karen-sparck-jones-iwd.html">Awesome People: Karen Spärck Jones</a>.<br />
<br />
<a href="http://engineering.columbia.edu/web/newsletter/files_engineeringnews/McKeown300.png" imageanchor="1" style="clear: right; float: right; margin-bottom: 1em; margin-left: 1em;"><img border="0" height="200" src="https://engineering.columbia.edu/web/newsletter/files_engineeringnews/McKeown300.png" width="150" /></a>Continuing on the topic of "who has been influential in my career and helped me get where I am?" today I'd like to talk about <a href="https://www.blogger.com/blogger.g?blogID=19803222">Kathy McKeown</a>, who is currently the Director of the Institute for Data Sciences and Engineering at Columbia. I had the pleasure of writing a mini-bio for Kathy for NAACL 2013 when I got to introduce her as one of the two invited speakers, and learned during that time that she was the first woman chair of computer science at Columbia and also the first woman to get tenure in the entirety of Columbia's School of Engineering and Applied Science. Kathy's name is near synonymous with ACL: she's held basically every elected position there is in our organization, in addition to being a AAAI, ACM, and ACL Fellow, and having won the Presidential Young Investigator Award from NSF, the Faculty Award for Women from NSF and the ABI Women of Vision Award.<br />
<br />
<div class="separator" style="clear: both; text-align: center;">
<a href="https://i.ytimg.com/vi/d-O_bFctgeY/hqdefault.jpg" imageanchor="1" style="clear: left; float: left; margin-bottom: 1em; margin-right: 1em;"><img border="0" height="150" src="https://i.ytimg.com/vi/d-O_bFctgeY/hqdefault.jpg" width="200" /></a></div>
One aspect of Kathy's research that I find really impressive is something that was highlighted in a nomination letter for her to be an invited speaker at NAACL 2013. I no longer have the original statement, but it was something like "Whenever a new topic becomes popular in NLP, we find out that Kathy worked on it ten years ago." This rings true of my own experience: recent forays into digital humanities, work on document and sentence compression, paraphrasing, technical term translation, and even her foundational work in the 80s on natural language interfaces to databases (now called "semantic parsing").<br />
<br />
Although---like Bonnie and Karen---I met Kathy through DUC as a graduate student, I didn't start working with her closely until I moved to Maryland and I had the opportunity to work on a big IARPA proposal with her as PI. That was the first of two really big proposals that she'd lead and I'd work on. These proposals involved both a huge amount of new-idea-generation and a huge amount of herding-professors, both of which are difficult in different ways.<br />
<br />
<div class="separator" style="clear: both; text-align: center;">
<a href="http://monod.biomath.nyu.edu/index/images/awis/awardees1.jpg" imageanchor="1" style="clear: right; float: right; margin-bottom: 1em; margin-left: 1em;"><img border="0" src="http://monod.biomath.nyu.edu/index/images/awis/awardees1.jpg" /></a></div>
On the research end, in the case of both proposals, Kathy's feedback on ideas has been invaluable. She's amazingly good at seeing through a convoluted idea and pushing on the parts that are either unclear or just plain don't make sense. She's really helped me hone my own ideas here.<br />
<br />
On the herding-professors end, I am so amazed with how Kathy manages a large team. We're currently having weekly phone calls, and one of the other co-PI's and I have observed in all seriousness that being on these phone calls is like free mentoring. I hope that one day I'd be able to manage even half of what Kathy manages.<br />
<br />
<div class="separator" style="clear: both; text-align: center;">
<a href="http://www8.gsb.columbia.edu/financialstudies/sites/financialstudies/files/Kathy2014.jpg" imageanchor="1" style="clear: left; float: left; margin-bottom: 1em; margin-right: 1em;"><img border="0" src="http://www8.gsb.columbia.edu/financialstudies/sites/financialstudies/files/Kathy2014.jpg" height="112" width="200" /></a></div>
One of my favorite less-research-y memories of Kathy was when our previous IARPA project was funded, she invited the entire team to a kickoff meeting in the Hamptons. It was the Fall, so the weather wasn't optimal, but a group of probably a ten faculty and twenty students converged there, ran around the beach, cooked dinner as a group, and bonded. And we discussed some research too. I still think back to this even regularly, because it's honestly not something I would have felt comfortable doing in her position. I have a tendency to keep my work life and my personal life pretty separate, and inviting thirty colleagues over for a kickoff meeting would've been way beyond my comfort zone: I think I worry about losing stature. Perhaps Kathy is more comfortable with this because of personality or because her stature is indisputable. Either way, it's made me think regularly about what sort of relationship I want and am comfortable with with students and colleagues.<br />
<br />
Spending any amount of time with Kathy is a learning experience for me, and I also have to thank Bonnie Dorr for including me on the first proposal with Kathy that kind of got me in the door. I'm incredibly indebted to her amazing intellect, impressive herding abilities, and open personality.<br />
<br />
Thanks, Kathy!halhttp://www.blogger.com/profile/02162908373916390369noreply@blogger.com0tag:blogger.com,1999:blog-19803222.post-23528637306134923512017-03-13T11:19:00.001-06:002017-03-13T11:21:08.511-06:00Awesome people: Karen Spärck Jones (IWD++)To honor women this <a href="https://www.internationalwomensday.com/">International Women's Day</a>, I have a several posts, broadly around the topic of women in STEM. Previous posts in this series include: <a href="https://nlpers.blogspot.com/2017/03/awesome-people-bonnie-dorr-iwd.html">Awesome People: Bonnie Dorr</a>, <a href="https://nlpers.blogspot.com/2017/03/awesome-people-ellen-riloff-iwd.html">Awesome People: Ellen Riloff</a> and <a href="https://nlpers.blogspot.com/2017/03/awesome-people-lise-getoor-iwd.html">Awesome People: Lise Getoor</a>.<br />
<br />
<a href="http://www.currybet.net/cbet_blog/2009/03/professor-karen-spark-jones--.php" style="clear: right; float: right; margin-bottom: 1em; margin-left: 1em;"><img src="http://www.currybet.net/images/blog2009/03/20090304_prof-spark-jones.jpg" height="200" width="163" /></a><br />
Today is the continuation of the theme "who has been influential in my career and helped me get where I am?" and in that vein, I want to talk about another awesome person: <a href="https://en.wikipedia.org/wiki/Karen_Sp%C3%A4rck_Jones">Karen Spärck Jones</a>. Like Bonnie Dorr, Karen is someone I first met at the Document Understanding Conference series back when I was a graduate student.<br />
<br />
Karen has done it all. First, she invents inverse document frequency, one of those topics that's so ingrained that no one even cites it anymore. I'm pretty sure I didn't know she invented IDF when I first met her. Frankly, I'm not sure it even occurred to me that this was something someone had to invent: it was like air or water. She's the <a href="http://www.cl.cam.ac.uk/misc/obituaries/sparck-jones/cv.html">recipient</a> of the AAAI Allen Newell Award, the BCS Lovelace Medal, the ACL lifetime achievement award, and ASIS&T Award of Merit, the Gerard Salton Award, was a fellow of the British Academy (and VP thereof), a fellow of AAAI, ECCAI and ACL. I highly recommend reading her speech from her <a href="http://www.mitpressjournals.org/doi/abs/10.1162/0891201053630237#.WLiaWldE6-k">ACL fellow award</a>. Among other things, I didn't realize that IDF was the fourth attempt to get the formulation right!<br />
<br />
If there are two things I learned from Karen, they are:
<br />
<ol>
<li> simple is good </li>
<li> examples are good </li>
</ol>
<a href="https://www.cl.cam.ac.uk/misc/obituaries/sparck-jones/video/" style="clear: left; float: left; margin-bottom: 1em; margin-right: 1em;"><img height="179" src="https://www.cl.cam.ac.uk/misc/obituaries/sparck-jones/video/award-lecture-2007.jpg" width="320" /></a>Although easily stated, these two principles are quite difficult to follow.I distinctly remember given a talk at DUC on <a href="https://www.blogger.com/blogger.g?blogID=19803222">BayeSum</a> and, afterward, Karen coming up to talk to me to try to get to the bottom of what the model was actually doing and why it was working, basically sure that there was a simpler explanation buried under the model.<br />
<br />
I also can't forget Karen routinely pushing people for examples in talks. Giving a talk on MT that doesn't have example outputs of your translation system? Better hope Karen isn't in the audience.<br />
<br />
Karen was also a huge proponent of breaking down gender barriers in computing. She's famously quoted as saying:<br />
<blockquote>
I think it's very important to get more women into computing. My slogan is: <b>"Computing is too important to be left to men."</b>
</blockquote>
This quote is a wonderful reflection both of Karen's seriousness and of her tongue-in-cheek humor. She was truly one of the kindest people I've met.<br />
<br />
<a href="http://savoyplace.theiet.org/engineering-hub/discover/gallery/index.cfm" style="clear: right; float: right; margin-bottom: 1em; margin-left: 1em;"><img src="http://savoyplace.theiet.org/engineering-hub/discover/gallery/img/engineers/images/thumbnails/JPEG/1935_Karen_Spaerck_Jones.jpg" /></a><br />
In particular, more than any of these specifics I just remember being so amazed and grateful that even as a third year graduate student, Karen, who was like this amazing figure in IR and summarization, would come talk to me for a half hour to help me make my research better. I was extremely sad nearly ten years ago when I learned that Karen has passed away. Just a week earlier, we had been exchanging emails about document collections, and the final email I had from her on the topic read as follows:<br />
<blockquote>
Document collections is a much misunderstood topic -- you have to think what its for and eg where (in retrieval) you are going to get real queries from. Just assembling some bunch of stuff and saying hey giys, what would you like to do with this is useless.
</blockquote>
This was true in 2007 and it's true today. In fact, I might argue that it's even more true today. We have nearly infinite ability to create datasets today, be them natural, artificial or whatever, and it's indeed not enough just to throw some stuff together and cross your fingers.<br />
<br />
I miss Karen so much. She had this joy that she brought everywhere and my life is less for that loss.
halhttp://www.blogger.com/profile/02162908373916390369noreply@blogger.com0tag:blogger.com,1999:blog-19803222.post-66365365788341344412017-03-10T05:46:00.000-07:002017-03-10T05:46:22.697-07:00Awesome people: Lise Getoor (IWD++)To honor women this <a href="https://www.internationalwomensday.com/">International Women's Day</a>, I have a several posts, broadly around the topic of women in STEM. Previous posts in this series include: <a href="https://nlpers.blogspot.com/2017/03/awesome-people-bonnie-dorr-iwd.html">Awesome People: Bonnie Dorr</a> and <a href="https://nlpers.blogspot.com/2017/03/awesome-people-ellen-riloff-iwd.html">Awesome People: Ellen Riloff</a>.<br />
<br />
Today is the continuation of the theme "who has been influential in my career and helped me get where I am?" and in that vein, I want to talk about another awesome person: Lise Getoor.<br />
<br />
<a href="https://www.soe.ucsc.edu/people/getoor" style="clear: right; float: right; margin-bottom: 1em; margin-left: 1em;"><img height="320" src="https://www.soe.ucsc.edu/people/getoor/photo/1" width="265" /></a>Lise is best known for her deep work in statistical relational learning, link mining, knowledge graph models and tons of applications to real world inference problems where data can be represented as a graph. She's currently a professor in CS at UCSC, but I had the fortune to spend a few years with her while she was still here at UMD. During this time, she was an NSF Career awardee, and is now a Fellow of the AAAI. At UMD when you're up for promotion, you give a "promotion talk" to the whole department, and I still remember sitting in her Full Prof promotion talk and being amazed---despite having known her for years at this point---at how well she made both deep technical contributions and also built software and tools that are useful for a huge variety of practitioners.<br />
<br />
Like Bonnie Dorr, Lise was something of an unofficial mentor to me. Ok I'll be honest. She's still an unofficial mentor to me. Faculty life is hard work, especially when one has just moved; going through tenure is stressful in a place where you haven't had years to learn how things work; none of which is made easier by simultaneously having personal-life challenges. Lise was always incredibly supportive in all of these areas, and I don't think I realized until after she had moved to UCSC how much I benefited from Lise's professional and emotional labor in helping me survive. And how helpful it is to have an openly supportive senior colleague to help grease some gears. I always felt like Lise was on my side.<br />
<br />
<a href="http://www.umiacs.umd.edu/~getoor/index.shtml" style="clear: left; float: left; margin-bottom: 1em; margin-right: 1em;"><img height="200" src="https://www.umiacs.umd.edu/~getoor/images/lcg.jpg" width="163" /></a>Probably one of the most important things I learned from Lise is how to be strategic, both in terms of research (what is actually worth putting my time and energy into) and departmental work (how can we best set ourselves up for success). As someone who has a tendency to spread himself too thin, it was incredibly useful to have a reminder that focusing on a smaller number of deeper things is more likely to have real lasting impact. I also found that I greatly respected her attention to excellence: my understanding (mostly from her students and postdocs) is that her personal acceptance rate on conference submissions is incredibly high (like almost 1.0), because her own internal bar for submission is generally much higher than any reviewer's. This is obviously something I haven't been able to replicate, but I think incredibly highly of Lise for this.<br />
<br />
Lise and I got promoted the same year---her to full prof, me to associate prof---and so we had a combined celebration dinner party at one of the (many) great Eritrean restaurants in DC followed by an attempt to go see life jazz at one of my favorite venues across the street. The music basically never showed up, but it was a really fun time anyway. Lise gave me a promotion gift that I still have on my desk: a small piece of wood (probably part of a branch of a tree) with a plaque that reads "Welcome to the World of Deadwood." This is particularly meaningful to me because Lise is so far from deadwood that it puts me to shame, and I can only hope to be as un-deadwood-like as her for the rest of my career.<br />
<br />
Thanks Lise!halhttp://www.blogger.com/profile/02162908373916390369noreply@blogger.com0tag:blogger.com,1999:blog-19803222.post-16534933259058414692017-03-09T07:50:00.000-07:002017-03-09T08:13:00.409-07:00Awesome people: Ellen Riloff (IWD++)To honor women this <a href="https://www.internationalwomensday.com/">International Women's Day</a>, I have a several posts, broadly around the topic of women in STEM. Previous posts in this series include: <a href="https://nlpers.blogspot.com/2017/03/awesome-people-bonnie-dorr-iwd.html">Awesome People: Bonnie Dorr</a>.<br />
<br />
<a href="https://www.cse.unt.edu/seminars/EllenRiloff.jpg" imageanchor="1" style="clear: right; float: right; margin-bottom: 1em; margin-left: 1em;"><img border="0" src="https://www.cse.unt.edu/seminars/EllenRiloff.jpg" /></a>Today is the continuation of the theme "who has been influential in my career and helped me get where I am?" and in that vein, I want to talk about <i>another</i> awesome person: <a href="http://www.cs.utah.edu/~riloff/">Ellen Riloff</a>. Ellen is a professor of computer science at the <a href="http://www.cs.utah.edu/">University of Utah</a>, and literally taught me everything I know about being a professor. I saw a joke a while ago that the transition from being a PhD student to a professor is like being trained for five years to swim and then being told to drive a boat. This was definitely true for me, and if it weren't for Ellen I'd have spent the past N years barely treading water. I truly appreciate the general inclusive and encouraging environment I belonged to during my time at Utah, and specifically appreciate everything Ellen did. When I think of Ellen as a researcher and as a person, I think: <i>honest</i> and <i>forthright</i>.<br />
<br />
Ellen is probably best known for her work on bootstrapping (for which she and Rosie Jones received a <a href="http://www.aaai.org/Awards/classic.php">AAAI Classic Paper award</a> in 2017) and information extraction (AAAI Classic Paper honorable mention in 2012), but has also worked more broadly on coreference resolution, sentiment analysis, active learning, and, in a wonderful project that also reveals <a href="http://www.cs.utah.edu/~riloff/personal.html">her profound love of animals</a>, <a href="http://www.aclweb.org/website/old_anthology/N/N15/N15-1168.pdf">veterinary medicine</a>. Although I only "officially" worked on one project with her (on <a href="http://www.aclweb.org/anthology/D/D10/D10-1008.pdf">plot units</a>), her influence on junior-faculty-Hal was deep and significant.<br />
<br />
I would be impossible to overstate how much impact Ellen has had on me as a researcher and a person. I still remember on my first NSF proposal, I sent her a draft and her comment main was "remove half." I was like "nooooooo!!!!" But she was right, and ever since them I try to repeat this advice to myself every time I write a proposal now.<br />
<br />
<a href="http://www.isi.edu/projects/sciknowmine/sciknowmine_release_workshop_-_bridging_bionlp_and_biocuration" imageanchor="1" style="clear: left; float: left; margin-bottom: 1em; margin-right: 1em;"><img border="0" src="https://i1.ytimg.com/vi/gHBsiEsw4_c/mqdefault.jpg" /></a>One of the most important scientific lessons I learned from Ellen is that how you construct your data matters. NLP is a field that's driven by the existence of data, but if we want NLP to be meaningful at all, we need to make sure that that data means what we think it means. Ellen's attention to detail in making sure that data was selected, annotated, and inspected correctly is deeper and more thoughtful than anyone else I've ever known. When we were working on the plot units stuff, we each spent about 30 minutes annotating a single fable, followed by another 30 minutes of adjudication, and then reannotation. I think we did about twenty of them. Could we have done it faster? Yes. Could we have had mechanical turkers do it? Probably. Would the data have been as meaningful? Of course not. Ellen taught me that when one releases a new dataset, this comes with a huge responsibility to make sure that it's carefully constructed and precise. Without that, the number of wasted hours of others that you run the risk of creating is huge. Whenever I work on building datasets these days, the Ellen-level-of-quality is my (often unreached) aspiration point.<br />
<br />
I distinctly remember a conversation Ellen and I had about advising Ph.D. students in my first or second year, in which I mentioned that I was having trouble figuring out how to motivate different students. Somewhat tongue-in-cheek, Ellen pointed out that different students are actually different people. Obvious (and amusing) in retrospect, but as I never saw my advisor interacting one on one with his other advisees, it actually had never occurred to me that he might have dealt with each of us differently. Like most new faculty, I also had to learn how to manage students, how to promote their work, how to correct them when they mis-step (because we all mis-step), and also how to do super important things like write letters. All of these things I learned from Ellen. I still try to follow her example as best I can.<br />
<br />
I was lucky enough to have the office just-next-door to Ellen, and we were both in our offices almost every weekday, and her openness to having me stick my head in her door to ask questions about anything from what are interesting grand research questions to how to handle issues with students, from how to write proposals to what do you want for lunch, was amazing. I feel like we had lunch together almost every day (that's probably an exaggeration, but that's how I remember it), and I owe her many thanks for helping me flesh out research ideas, and generally function as a junior faculty member. She was without a doubt the single biggest impact on my life as junior faculty, and I remain deeply indebted to her for everything she did directly and behind the scenes.<br />
<br />
Thanks Ellen!
halhttp://www.blogger.com/profile/02162908373916390369noreply@blogger.com2tag:blogger.com,1999:blog-19803222.post-66437072349218676052017-03-08T08:34:00.001-07:002017-03-08T08:47:01.249-07:00Awesome people: Bonnie Dorr (IWD++)To honor women this <a href="https://www.internationalwomensday.com/">International Women's Day</a>, I have a several posts, broadly around the topic of women in STEM.<br />
<br />
This is the first, and the topic is "who has been influential in my career and helped me get where I am?" There are many such people, and any list will be woefully incomplete, but today I'm going to highlight <a href="https://www.ihmc.us/groups/bdorr/">Bonnie Dorr</a> (who founded the <a href="https://wiki.umiacs.umd.edu/clip/index.php/Main_Page">CLIP lab</a> together with <a href="http://www.umiacs.umd.edu/~weinberg/">Amy Weinberg</a> and <a href="http://www.umiacs.umd.edu/users/louiqa/">Louiqa Raschid</a>, and who also is a <a href="https://www.cs.umd.edu/article/2016/11/emerita-professor-bonnie-dorr-named-acl-fellow-2016">recent fellow of the ACL</a>!).<br />
<br />
<a href="https://www.umiacs.umd.edu/people/bonnie" imageanchor="1" style="clear: right; float: right; margin-bottom: 1em; margin-left: 1em;"><img border="0" src="https://www.umiacs.umd.edu/sites/default/files/styles/medium/public/dorr.png?itok=BLPUBgrD" /></a>For those who haven't had the chance to work with Bonnie, you're missing out. I don't know how she does it, but the depth and speed at which she interacts, works, produces ideas and gets things done is stunning. Before leaving for a program manager position at DARPA and then later to IHMC, Bonnie was full professor (and then associate dean) here at UMD. At DARPA she managed basically two PM's worth of projects, and was always excited about everything. During her time as a professor here at UMD (after earning her Ph.D. from MIT), Bonnie was an NSF Presidential Faculty Fellow, a Sloan Recipient, a recipient of the NSF Young Investigator Award, and a AAAI Fellow.<br />
<br />
I learned a lot from Bonnie. I first met her back when I was a graduate student and Daniel and I had a paper in the Document Understanding Conference (basically the summarization workshop of the day) on evaluation. It was closely related to something Bonnie had worked on previously, and I was really thrilled to get feedback from her. Fast forward six years and then I'm writing proposals with Bonnie, advising postdocs and students together, and otherwise trying to learn as much as possible by osmosis and direct instruction.<br />
<br />
<a href="http://www.cs.umd.edu/local-cgi-bin/csphotohistory/description.php?lname=Dorr&fname=Bonnie%20J.&Mainpic=bonniedorr.jpg" imageanchor="1" style="clear: left; float: left; margin-bottom: 1em; margin-right: 1em;"><img border="0" height="217" src="https://www.cs.umd.edu/projects/photohistory/facultypictures_full/bonniedorr.jpg" width="320" /></a>One of the most important things I learned from Bonnie was: if you want it done, just do it. Bonnie is a do-er. This is reflected in her incredibly broad scientific contributions (summarization, machine translation, evaluation, etc.) as well as the impact she had on the department. It was clear almost immediately that the faculty here really respected Bonnie's opinion; her ability to move mountains was evident.<br />
<br />
On a more personal note, although she was not my official senior-faculty-mentor when I came to UMD, Bonnie was one of two senior faculty members here who really did everything she could to help me---both professionally and personally. Whenever I was on the fence about how to handle something, I knew that I could go to Bonnie and get her opinion and that her opinion would be well reasoned. I wouldn't always take it (sometimes to my own chagrin), but she was always ready with concrete advice about specific steps to take about almost any topic. I've also been on two very-large grant proposals with her (one successful and one not) which have both been incredible learning experiences. Getting a dozen faculty to work on a 30 page document is no easy task, and Bonnie's combination of just-do-it and lead-by-example is something I still try to mimic when I'm in a similar (if smaller) position. Even when she was at DARPA, as well as now, as professor emerita at UMD, she's still actively supporting both me and other faculty here, and clearly really cares that people at UMD are successful.<br />
<br />
<a href="http://www.cs.umd.edu/local-cgi-bin/csphotohistory/description.php?lname=Dorr&fname=Bonnie%20J.&Mainpic=bonniedorr.jpg" imageanchor="1" style="clear: right; float: right; margin-bottom: 1em; margin-left: 1em;"><img border="0" height="240" src="https://www.umiacs.umd.edu/~bonnie/CLIP_In_Waikiki_2008-Take2.jpg" width="320" /></a>In addition to Bonnie's seriousness and excellence in research and professional life, I also really appreciated her more laid back side. When I visited UMD back before accepting a job here, she hosted a visit day dinner for prospective grad students at her house, which overlapped with one of her student's Ph.D. defense: hence, a combined party. To honor the student, Bonnie had written a rap, which she then performed with her son beatboxing. It was in that moment that I realized truly how amazing Bonnie is not just as a researcher but as a person. (Of course, she attacked this task with exactly the same high intensity that she attacks every other problem!)<br />
<br />
Overall, Bonnie is both one of the most amazing researchers I know, one of the strongest go-getters I know, and someone I've been extremely luck to have not just as a collaborator, but also a colleague and mentor.<br />
<br />
Thanks Bonnie!
<br />
<br />halhttp://www.blogger.com/profile/02162908373916390369noreply@blogger.com1tag:blogger.com,1999:blog-19803222.post-80466321468595096752016-12-12T07:39:00.001-07:002016-12-13T14:27:51.485-07:00Should the NLP and ML Communities have a Code of Ethics?At ACL this past summer, Dirk Hovy and Shannon Spruit presented a very nice opinion paper on Ethics in NLP. There's also been a great surge of interest in FAT-everything (FAT = Fairness, Accountability and Transparency), typified by FATML, but there are others. And yet, despite this recent interest in ethics-related topics, none of the major organizations that I'm involved in have a Code of Ethics, namely: the <a href="http://aclweb.org/">ACL</a>, the <a href="https://nips.cc/About">NIPS foundation</a> nor the <a href="http://www.machinelearning.org/">IMLS</a>. After Dirk's presentation, he, I, <a href="http://www.m-mitchell.com/">Meg Mitchell</a> and <a href="http://michael.kimstrube.de/">Michael Strube</a> and some others spent a while discussing ethics in NLP (a different subset, including Dirk, Shannon, Meg, Hanna Wallach, Michael and Emily Bender) went on to form a workshop that'll take place at EACL) and the possibility of a code of ethics (Meg also brought this up at the ACL business meeting) which eventually gave rise to this post.<br />
<br />
<span style="font-size: x-small;">(Note: the NIPS foundation has a "Code of Conduct" that I hadn't seen before that covers similar ground to the (NA)ACL <a href="http://naacl.org/policies/anti-harassment.html">Anti-Harassment Policy</a>; what I'm talking about here is different.)</span><br />
<br />
A natural question is whether this is unusual among professional organizations. The answer is most definitely <b>yes, it's very unusual.</b> The <a href="http://acm.org/">Association for Computing Machinery</a>, the <a href="http://bcs.org/">British Computer Society</a>, the <a href="http://ieee.org/">IEEE</a>, the <a href="http://www.linguisticsociety.org/">Linguistic Society of America</a> and the <a href="http://www.ciol.org.uk/">Chartered Institute of Linguistics</a> all have codes of ethics (or codes of conduct). In most cases, agreeing to the code of ethics is a prerequisite to membership in these communities.<br />
<br />
There is a nice set of UMD EE slides on <a href="https://www.ece.umd.edu/Academic/ethics/overheads.pdf">Ethics in Engineering</a> that describes why one might want a code of Ethics:<br />
<ol>
<li>Provides a framework for ethical judgments within a profession</li>
<li>Expresses the commitment to shared minimum standards for acceptable behavior</li>
<li>Provides support and guidance for those seeking to act ethically</li>
<li>Formal basis for investigating unethical conduct, as such, may serve both as a deterrence to and discipline for unethical behavior</li>
</ol>
Personally, I've never been a member of any of these societies that have codes of ethics. Each organization has a different set of codes, largely because they need to address issues specific to their field of expertise. For instance, linguists often do field work, and in doing so often interact with indigenous populations.<br />
<br />
Below, I have reproduced the <a href="http://www.ieee.org/about/corporate/governance/p7-8.html">IEEE code</a> as a representative sample (emphasis mine) because it is relatively brief. The <a href="https://www.acm.org/about-acm/acm-code-of-ethics-and-professional-conduct">ACM code</a> and <a href="http://www.bcs.org/upload/pdf/conduct.pdf">BCS code</a> are slightly different, and go into more details. The <a href="http://www.linguisticsociety.org/resource/ethics">LSA code</a> and <a href="http://www.ciol.org.uk/code">CIL code</a> are related but cover slightly different topics. By being a member of the IEEE, one agrees:<br />
<ol>
<li>to accept responsibility in <span style="background-color: #fff2cc;"><span style="color: blue;"><b>making decisions consistent with the
safety, health, and welfare of the public</b></span></span>, and to disclose promptly
factors that might endanger the public or the environment;</li>
<li>to avoid real or perceived conflicts of interest whenever possible, and to disclose them to affected parties when they do exist;</li>
<li><span style="background-color: #fff2cc;"><span style="color: blue;"><b>to be honest and realistic in stating claims</b></span></span> or estimates based on available data; </li>
<li>to reject bribery in all its forms; </li>
<li><span style="background-color: #fff2cc;"><span style="color: blue;"><b>to improve the understanding of technology; its appropriate application, and potential consequences</b></span></span>; </li>
<li>to
maintain and improve our technical competence and to undertake
technological tasks for others only if qualified by training or
experience, or after full disclosure of pertinent limitations; </li>
<li>to
seek, accept, and offer honest criticism of technical work, to
acknowledge and correct errors, and to credit properly the contributions
of others; </li>
<li>to treat fairly all persons and to not engage in
acts of discrimination based on race, religion, gender, disability, age,
national origin, sexual orientation, gender identity, or gender
expression;</li>
<li>to avoid injuring others, their property, reputation, or employment by false or malicious action; </li>
<li>to assist colleagues and co-workers in their professional development and to support them in following this code of ethics.</li>
</ol>
The pieces I've highlighted are things that I think are especially important to think about, and places where I think we, as a community, might need to work harder.<br />
<br />
After this past ACL, I spent some time combing through the Codes of Ethics mentioned before and tried to synthesize a list that would make sense for the ACL, IMLS or NIPS. This is in very "drafty" form, but hopefully the content makes sense. Also to be 100% clear, <i>all of this is basically copy and paste with minor edits from one or more of the Codes linked above; nothing here is original.</i><br />
<br />
<ol id="docs-internal-guid-de5d562f-e97b-9a31-0364-824de62d2953" style="margin-bottom: 0pt; margin-top: 0pt;">
<li dir="ltr" style="background-color: transparent; color: black; font-family: Arial; font-size: 14.666666666666666px; font-style: normal; font-variant: normal; font-weight: 400; list-style-type: decimal; text-decoration: none; vertical-align: baseline;"><div dir="ltr" style="line-height: 1.38; margin-bottom: 0pt; margin-top: 0pt;">
<span style="background-color: transparent; color: black; font-family: "arial"; font-size: 14.666666666666666px; font-style: normal; font-variant: normal; font-weight: 400; text-decoration: none; vertical-align: baseline;">Responsibility to the Public</span></div>
</li>
<ol style="margin-bottom: 0pt; margin-top: 0pt;">
<li dir="ltr" style="background-color: transparent; color: black; font-family: Arial; font-size: 14.666666666666666px; font-style: normal; font-variant: normal; font-weight: 400; list-style-type: lower-alpha; text-decoration: none; vertical-align: baseline;"><div dir="ltr" style="line-height: 1.38; margin-bottom: 0pt; margin-top: 0pt;">
<span style="background-color: transparent; color: black; font-family: "arial"; font-size: 14.666666666666666px; font-style: normal; font-variant: normal; font-weight: 400; text-decoration: none; vertical-align: baseline;">Make research available to general public</span></div>
</li>
<li dir="ltr" style="background-color: transparent; color: black; font-family: Arial; font-size: 14.666666666666666px; font-style: normal; font-variant: normal; font-weight: 400; list-style-type: lower-alpha; text-decoration: none; vertical-align: baseline;"><div dir="ltr" style="line-height: 1.38; margin-bottom: 0pt; margin-top: 0pt;">
<span style="background-color: transparent; color: black; font-family: "arial"; font-size: 14.666666666666666px; font-style: normal; font-variant: normal; font-weight: 400; text-decoration: none; vertical-align: baseline;">Be honest and realistic in stating claims; ensure empirical bases and limitations are communicated appropriately</span></div>
</li>
<li dir="ltr" style="background-color: transparent; color: black; font-family: Arial; font-size: 14.666666666666666px; font-style: normal; font-variant: normal; font-weight: 400; list-style-type: lower-alpha; text-decoration: none; vertical-align: baseline;"><div dir="ltr" style="line-height: 1.38; margin-bottom: 0pt; margin-top: 0pt;">
<span style="background-color: transparent; color: black; font-family: "arial"; font-size: 14.666666666666666px; font-style: normal; font-variant: normal; font-weight: 400; text-decoration: none; vertical-align: baseline;">Only accept work and make statements on topics which you believe have competence to do</span></div>
</li>
<li dir="ltr" style="background-color: transparent; color: black; font-family: Arial; font-size: 14.666666666666666px; font-style: normal; font-variant: normal; font-weight: 400; list-style-type: lower-alpha; text-decoration: none; vertical-align: baseline;"><div dir="ltr" style="line-height: 1.38; margin-bottom: 0pt; margin-top: 0pt;">
<span style="background-color: transparent; color: black; font-family: "arial"; font-size: 14.666666666666666px; font-style: normal; font-variant: normal; font-weight: 400; text-decoration: none; vertical-align: baseline;">Contribute to society and human well-being, and minimize negative consequences of computing systems</span></div>
</li>
<li dir="ltr" style="background-color: transparent; color: black; font-family: Arial; font-size: 14.666666666666666px; font-style: normal; font-variant: normal; font-weight: 400; list-style-type: lower-alpha; text-decoration: none; vertical-align: baseline;"><div dir="ltr" style="line-height: 1.38; margin-bottom: 0pt; margin-top: 0pt;">
<span style="background-color: transparent; color: black; font-family: "arial"; font-size: 14.666666666666666px; font-style: normal; font-variant: normal; font-weight: 400; text-decoration: none; vertical-align: baseline;">Make reasonable effort to prevent misinterpretation of results</span></div>
</li>
<li dir="ltr" style="background-color: transparent; color: black; font-family: Arial; font-size: 14.666666666666666px; font-style: normal; font-variant: normal; font-weight: 400; list-style-type: lower-alpha; text-decoration: none; vertical-align: baseline;"><div dir="ltr" style="line-height: 1.38; margin-bottom: 0pt; margin-top: 0pt;">
<span style="background-color: transparent; color: black; font-family: "arial"; font-size: 14.666666666666666px; font-style: normal; font-variant: normal; font-weight: 400; text-decoration: none; vertical-align: baseline;">Make decisions consistent with safety, health & welfare of public</span></div>
</li>
<li dir="ltr" style="background-color: transparent; color: black; font-family: Arial; font-size: 14.666666666666666px; font-style: normal; font-variant: normal; font-weight: 400; list-style-type: lower-alpha; text-decoration: none; vertical-align: baseline;"><div dir="ltr" style="line-height: 1.38; margin-bottom: 0pt; margin-top: 0pt;">
<span style="background-color: transparent; color: black; font-family: "arial"; font-size: 14.666666666666666px; font-style: normal; font-variant: normal; font-weight: 400; text-decoration: none; vertical-align: baseline;">Improve understanding of technology, its application and its potential consequences (positive and negative)</span></div>
</li>
</ol>
<li dir="ltr" style="background-color: transparent; color: black; font-family: Arial; font-size: 14.666666666666666px; font-style: normal; font-variant: normal; font-weight: 400; list-style-type: decimal; text-decoration: none; vertical-align: baseline;"><div dir="ltr" style="line-height: 1.38; margin-bottom: 0pt; margin-top: 0pt;">
<span style="background-color: transparent; color: black; font-family: "arial"; font-size: 14.666666666666666px; font-style: normal; font-variant: normal; font-weight: 400; text-decoration: none; vertical-align: baseline;">Responsibility in Research</span></div>
</li>
<ol style="margin-bottom: 0pt; margin-top: 0pt;">
<li dir="ltr" style="background-color: transparent; color: black; font-family: Arial; font-size: 14.666666666666666px; font-style: normal; font-variant: normal; font-weight: 400; list-style-type: lower-alpha; text-decoration: none; vertical-align: baseline;"><div dir="ltr" style="line-height: 1.38; margin-bottom: 0pt; margin-top: 0pt;">
<span style="background-color: transparent; color: black; font-family: "arial"; font-size: 14.666666666666666px; font-style: normal; font-variant: normal; font-weight: 400; text-decoration: none; vertical-align: baseline;">Protect the personal identification of research subjects, and abide by informed consent</span></div>
</li>
<li dir="ltr" style="background-color: transparent; color: black; font-family: Arial; font-size: 14.666666666666666px; font-style: normal; font-variant: normal; font-weight: 400; list-style-type: lower-alpha; text-decoration: none; vertical-align: baseline;"><div dir="ltr" style="line-height: 1.38; margin-bottom: 0pt; margin-top: 0pt;">
<span style="background-color: transparent; color: black; font-family: "arial"; font-size: 14.666666666666666px; font-style: normal; font-variant: normal; font-weight: 400; text-decoration: none; vertical-align: baseline;">Conduct research honestly, avoiding plagiarism and fabrication of results</span></div>
</li>
<li dir="ltr" style="background-color: transparent; color: black; font-family: Arial; font-size: 14.666666666666666px; font-style: normal; font-variant: normal; font-weight: 400; list-style-type: lower-alpha; text-decoration: none; vertical-align: baseline;"><div dir="ltr" style="line-height: 1.38; margin-bottom: 0pt; margin-top: 0pt;">
<span style="background-color: transparent; color: black; font-family: "arial"; font-size: 14.666666666666666px; font-style: normal; font-variant: normal; font-weight: 400; text-decoration: none; vertical-align: baseline;">Cite prior work as appropriate</span></div>
</li>
<li dir="ltr" style="background-color: transparent; color: black; font-family: Arial; font-size: 14.666666666666666px; font-style: normal; font-variant: normal; font-weight: 400; list-style-type: lower-alpha; text-decoration: none; vertical-align: baseline;"><div dir="ltr" style="line-height: 1.38; margin-bottom: 0pt; margin-top: 0pt;">
<span style="background-color: transparent; color: black; font-family: "arial"; font-size: 14.666666666666666px; font-style: normal; font-variant: normal; font-weight: 400; text-decoration: none; vertical-align: baseline;">Preserve original data and documentation, and make available</span></div>
</li>
<li dir="ltr" style="background-color: transparent; color: black; font-family: Arial; font-size: 14.666666666666666px; font-style: normal; font-variant: normal; font-weight: 400; list-style-type: lower-alpha; text-decoration: none; vertical-align: baseline;"><div dir="ltr" style="line-height: 1.38; margin-bottom: 0pt; margin-top: 0pt;">
<span style="background-color: transparent; color: black; font-family: "arial"; font-size: 14.666666666666666px; font-style: normal; font-variant: normal; font-weight: 400; text-decoration: none; vertical-align: baseline;">Follow through on promises made in grant proposals and acknowledge support of sponsors</span></div>
</li>
<li dir="ltr" style="background-color: transparent; color: black; font-family: Arial; font-size: 14.666666666666666px; font-style: normal; font-variant: normal; font-weight: 400; list-style-type: lower-alpha; text-decoration: none; vertical-align: baseline;"><div dir="ltr" style="line-height: 1.38; margin-bottom: 0pt; margin-top: 0pt;">
<span style="background-color: transparent; color: black; font-family: "arial"; font-size: 14.666666666666666px; font-style: normal; font-variant: normal; font-weight: 400; text-decoration: none; vertical-align: baseline;">Avoid real or perceived COIs, disclose when they exist; reject bribery</span></div>
</li>
<li dir="ltr" style="background-color: transparent; color: black; font-family: Arial; font-size: 14.666666666666666px; font-style: normal; font-variant: normal; font-weight: 400; list-style-type: lower-alpha; text-decoration: none; vertical-align: baseline;"><div dir="ltr" style="line-height: 1.38; margin-bottom: 0pt; margin-top: 0pt;">
<span style="background-color: transparent; color: black; font-family: "arial"; font-size: 14.666666666666666px; font-style: normal; font-variant: normal; font-weight: 400; text-decoration: none; vertical-align: baseline;">Honor property rights, including copyrights and patents</span></div>
</li>
<li dir="ltr" style="background-color: transparent; color: black; font-family: Arial; font-size: 14.666666666666666px; font-style: normal; font-variant: normal; font-weight: 400; list-style-type: lower-alpha; text-decoration: none; vertical-align: baseline;"><div dir="ltr" style="line-height: 1.38; margin-bottom: 0pt; margin-top: 0pt;">
<span style="background-color: transparent; color: black; font-family: "arial"; font-size: 14.666666666666666px; font-style: normal; font-variant: normal; font-weight: 400; text-decoration: none; vertical-align: baseline;">Seek, accept and offer honest criticism of technical work; correct errors; provide appropriate professional review</span></div>
</li>
</ol>
<li dir="ltr" style="background-color: transparent; color: black; font-family: Arial; font-size: 14.666666666666666px; font-style: normal; font-variant: normal; font-weight: 400; list-style-type: decimal; text-decoration: none; vertical-align: baseline;"><div dir="ltr" style="line-height: 1.38; margin-bottom: 0pt; margin-top: 0pt;">
<span style="background-color: transparent; color: black; font-family: "arial"; font-size: 14.666666666666666px; font-style: normal; font-variant: normal; font-weight: 400; text-decoration: none; vertical-align: baseline;">Responsibility to Students, Colleagues, and other Researchers</span></div>
</li>
<ol style="margin-bottom: 0pt; margin-top: 0pt;">
<li dir="ltr" style="background-color: transparent; color: black; font-family: Arial; font-size: 14.666666666666666px; font-style: normal; font-variant: normal; font-weight: 400; list-style-type: lower-alpha; text-decoration: none; vertical-align: baseline;"><div dir="ltr" style="line-height: 1.38; margin-bottom: 0pt; margin-top: 0pt;">
<span style="background-color: transparent; color: black; font-family: "arial"; font-size: 14.666666666666666px; font-style: normal; font-variant: normal; font-weight: 400; text-decoration: none; vertical-align: baseline;">Recognize and property attribute contributions of students; promote student contributions to research</span></div>
</li>
<li dir="ltr" style="background-color: transparent; color: black; font-family: Arial; font-size: 14.666666666666666px; font-style: normal; font-variant: normal; font-weight: 400; list-style-type: lower-alpha; text-decoration: none; vertical-align: baseline;"><div dir="ltr" style="line-height: 1.38; margin-bottom: 0pt; margin-top: 0pt;">
<span style="background-color: transparent; color: black; font-family: "arial"; font-size: 14.666666666666666px; font-style: normal; font-variant: normal; font-weight: 400; text-decoration: none; vertical-align: baseline;">No discrimination based on gender identity, gender expression, disability, marital status, race/ethnicity, class, politics, religion, national origin, sexual orientation, age, etc. (should this go elsewhere?)</span></div>
</li>
<li dir="ltr" style="background-color: transparent; color: black; font-family: Arial; font-size: 14.666666666666666px; font-style: normal; font-variant: normal; font-weight: 400; list-style-type: lower-alpha; text-decoration: none; vertical-align: baseline;"><div dir="ltr" style="line-height: 1.38; margin-bottom: 0pt; margin-top: 0pt;">
<span style="background-color: transparent; color: black; font-family: "arial"; font-size: 14.666666666666666px; font-style: normal; font-variant: normal; font-weight: 400; text-decoration: none; vertical-align: baseline;">Teach students ethical responsibilities</span></div>
</li>
<li dir="ltr" style="background-color: transparent; color: black; font-family: Arial; font-size: 14.666666666666666px; font-style: normal; font-variant: normal; font-weight: 400; list-style-type: lower-alpha; text-decoration: none; vertical-align: baseline;"><div dir="ltr" style="line-height: 1.38; margin-bottom: 0pt; margin-top: 0pt;">
<span style="background-color: transparent; color: black; font-family: "arial"; font-size: 14.666666666666666px; font-style: normal; font-variant: normal; font-weight: 400; text-decoration: none; vertical-align: baseline;">Avoid injuring others, their property, reputation or employment by false or malicious action</span></div>
</li>
<li dir="ltr" style="background-color: transparent; color: black; font-family: Arial; font-size: 14.666666666666666px; font-style: normal; font-variant: normal; font-weight: 400; list-style-type: lower-alpha; text-decoration: none; vertical-align: baseline;"><div dir="ltr" style="line-height: 1.38; margin-bottom: 0pt; margin-top: 0pt;">
<span style="background-color: transparent; color: black; font-family: "arial"; font-size: 14.666666666666666px; font-style: normal; font-variant: normal; font-weight: 400; text-decoration: none; vertical-align: baseline;">Respect the privacy of others and honor confidentiality</span></div>
</li>
<li dir="ltr" style="background-color: transparent; color: black; font-family: Arial; font-size: 14.666666666666666px; font-style: normal; font-variant: normal; font-weight: 400; list-style-type: lower-alpha; text-decoration: none; vertical-align: baseline;"><div dir="ltr" style="line-height: 1.38; margin-bottom: 0pt; margin-top: 0pt;">
<span style="background-color: transparent; color: black; font-family: "arial"; font-size: 14.666666666666666px; font-style: normal; font-variant: normal; font-weight: 400; text-decoration: none; vertical-align: baseline;">Honor contracts, agreements and assigned responsibilities</span></div>
</li>
</ol>
<li dir="ltr" style="background-color: transparent; color: black; font-family: Arial; font-size: 14.666666666666666px; font-style: normal; font-variant: normal; font-weight: 400; list-style-type: decimal; text-decoration: none; vertical-align: baseline;"><div dir="ltr" style="line-height: 1.38; margin-bottom: 0pt; margin-top: 0pt;">
<span style="background-color: transparent; color: black; font-family: "arial"; font-size: 14.666666666666666px; font-style: normal; font-variant: normal; font-weight: 400; text-decoration: none; vertical-align: baseline;">Compliance with the code</span></div>
</li>
<ol style="margin-bottom: 0pt; margin-top: 0pt;">
<li dir="ltr" style="background-color: transparent; color: black; font-family: Arial; font-size: 14.666666666666666px; font-style: normal; font-variant: normal; font-weight: 400; list-style-type: lower-alpha; text-decoration: none; vertical-align: baseline;"><div dir="ltr" style="line-height: 1.38; margin-bottom: 0pt; margin-top: 0pt;">
<span style="background-color: transparent; color: black; font-family: "arial"; font-size: 14.666666666666666px; font-style: normal; font-variant: normal; font-weight: 400; text-decoration: none; vertical-align: baseline;">Uphold and promote the principles of this code</span></div>
</li>
<li dir="ltr" style="background-color: transparent; color: black; font-family: Arial; font-size: 14.666666666666666px; font-style: normal; font-variant: normal; font-weight: 400; list-style-type: lower-alpha; text-decoration: none; vertical-align: baseline;"><div dir="ltr" style="line-height: 1.38; margin-bottom: 0pt; margin-top: 0pt;">
<span style="background-color: transparent; color: black; font-family: "arial"; font-size: 14.666666666666666px; font-style: normal; font-variant: normal; font-weight: 400; text-decoration: none; vertical-align: baseline;">Treat violations of this code as inconsistent with membership in this organization </span></div>
<div dir="ltr" style="line-height: 1.38; margin-bottom: 0pt; margin-top: 0pt;">
<span style="background-color: transparent; color: black; font-family: "arial"; font-size: 14.666666666666666px; font-style: normal; font-variant: normal; font-weight: 400; text-decoration: none; vertical-align: baseline;"><br /></span></div>
</li>
</ol>
</ol>
I'd love to see (and help, if wanted) ACL, IMLS and NIPS foundation work on constructing a code of ethics. Our fields are more and more dealing with problems that have real impact on society, and I would like to see us, as a community, come together and express our shared standards.<br />
<br />halhttp://www.blogger.com/profile/02162908373916390369noreply@blogger.com7tag:blogger.com,1999:blog-19803222.post-33238481955074924142016-12-09T08:44:00.001-07:002016-12-09T08:44:05.092-07:00Whence your reward function?I ran a grad seminar in <a href="http://ter.ps/rl16">reinforcement learning this past semester</a>, which was a lot of fun and also gave me an opportunity to catch up on some stuff I'd been meaning to learn but haven't had a chance and old stuff I'd largely forgotten about. It's hard to believe, but my first RL paper was eleven years ago at a NIPS workshop where Daniel Marcu, John Langford and I had a <a href="http://www.umiacs.umd.edu/~hal/docs/daume05search.pdf">first paper on reducing structured prediction to reinforcement learning</a>, essentially by running <a href="http://hal3.name/courses/2016F_RL/Kakade02.pdf">Conservative Policy Iteration</a>. (This work eventually became <a href="http://hal3.name/docs/daume09searn.pdf">Searn</a>.) Most of my own work in the RL space has focused on imitation learning/learning from demonstrations, but me and my students have recently been pushing more into straight up reinforcement learning <a href="http://hal3.name/docs/daume15lols.pdf">algorithms</a> and <a href="http://www.umiacs.umd.edu/~hhe/papers/2016_icml_opponent.pdf">applications</a> and <a href="http://arxiv.org/abs/1602.02181">explanations</a> (also see <a href="https://people.csail.mit.edu/taolei/papers/emnlp16_rationale.pdf">Tao Lei's awesome work in a similar explanations vein</a>, and Tim Vieira's really nice <a href="https://timvieira.github.io/doc/2016-tacl-pruning.pdf">TACL paper</a> too).<br />
<br />
Reinforcement learning has undergone a bit of a renaissance recently, largely due to the efficacy of its combination with good function approximation via deep neural networks. Even more arguably this advance has been due to the increased availability and interest in "interesting" simulated environments, mostly video games and typified by the Atari game collection. In a very similar way that ImageNet made neural networks really work for computer vision (by being large, and capitalizing on the existence of GPUs), I think it's fair to say that these simulated environments have provided the same large data setting for RL that can also be combined with GPU power to build impressive solutions to many games.<br />
<br />
In a real sense, many parts of the RL community are going all-in on the notion that learning to play games is a path toward broader AI. The usual refrain that I hear arguing <i>against</i> that approach is based on the quantity of data. The argument is roughly: if you actually want to build a robot that acts in the real world, you're not going to be able to simulate 10 million frames (from the <a href="https://deepmind.com/research/dqn/">Deepmind paper</a>, which is just under 8 days of real time experience).<br />
<br />
I think this is an issue, but I actually don't think it's the most substantial issue. I think the most substantial issue is the fact that game playing is a simulated environment and the reward function is generally crafted to make humans find the games fun, which usually means frequent small rewards that point you in the right direction. This is exactly where RL works well, and something that I'm not sure is a reasonable assumption in the real world.<br />
<br />
Delayed reward is one of the hardest issues in RL, because (a) it means you have to do a lot of exploration and (b) you have a significant credit assignment problem. For instance, if you imagine a variant of (pick your favorite video game) where you only get a +1/-1 reward at the end of the game that says whether you won or lost, it becomes much much harder to learn, even if you play 10 million frames or 10 billion frames.<br />
<br />
That's all to say: games are really nice settings for RL because there's a very well defined reward function and you typically get that reward very frequently. Neither of these things is going to be true in the real world, regardless of how much data you have.<br /><br />At the end of the day, playing video games, while impressive, is really not that different from doing classification on synthetic data. Somehow it's better because the people doing the research were not those who invented the synthetic data, but games---even recent games that you might play on your (insert whatever the current popular gaming system is) are still heavily designed---are built in such a way that they are fun for their human players, which typically means increasing difficulty/complexity and relatively regularly reward function.<br /><br />As we move toward systems that we expect to work in the real world (even if that is not embodied---I don't necessarily mean the difficulty of physical robots), it's less and less clear where the reward function comes from.<br /><br />One option is to design a reward function. For complex behavior, I don't think we have any idea how to do this. There is the joke example in the R+N AI textbook where you give a vacuum cleaner a reward function for number of pieces of gunk picked up; the vacuum learns to pick up gunk, then drop it, then pick it up again, ad infinitum. It's a silly example, but I don't think we have much of an understanding of how to design reward functions for truly complex behaviors without significant risk of "unintended consequences." (To point a finger toward myself, we invented a reward function for simultaneous interpretation called <a href="http://hal3.name/docs/daume14simultaneousmt.pdf">Latency-Bleu</a> a while ago, and six months later we realized there's a very simple way to game this metric. I was then disappointed that the models never learned that exploit.)<br /><br />This is one reason I've spent most of my RL effort on imitation learning (IL) like things, typically where you can simulate an oracle. I've rarely seen an NLP problem that's been solved with RL where I haven't thought that it would have been much better and easier to just do IL. Of course IL has it's own issues: it's not a panacea.<br /><br />One thing I've been thinking about a lot recently is forms of implicit feedback. One cool paper in this area I learned about when I visited GATech a few weeks ago is <a href="http://ieeexplore.ieee.org/abstract/document/7742965/?reload=true">Learning from Explanations using Sentiment and Advice in RL</a> by Samantha Krening and colleagues. In this work they basically have a coach sitting on the side of an RL algorithm giving it advice, and used that to tailor things that I think of as more immediate reward. I generally think of this kind of like a baby. There's some built in reward signal (it can't be turtles all the way down), but what we think of as a reward signal (like a friend saying "I really don't like that you did that") only turn into this true reward through a learned model that tells me that that's negative feedback. I'd love to see more work in the area of trying to figure out how to transform sparse and imperfect "true" reward signals into something that we can actually learn to optimize.<br />halhttp://www.blogger.com/profile/02162908373916390369noreply@blogger.com3tag:blogger.com,1999:blog-19803222.post-3714575973895909872016-11-28T08:31:00.004-07:002016-11-28T09:38:46.101-07:00Workshops and mini-conferencesI've attended and organized two types of workshops in my time, one of which I'll call the ACL-style workshop (or "mini-conference"), the other of which I'll call the NIPS-style workshop (or "actual workshop"). Of course this is a continuum, and some workshops at NIPS are ACL-style and vice versa. As I've already given away with phrasing, I much prefer the NIPS style. Since many NLPers may never have been to NIPS or to a NIPS workshop, I'm going to try to describe the differences and explain my preference (and also highlight some difficulties).<br />
<br />
(Note: when I say ACL-style workshop, I'm <i>not</i> thinking of things like WMT that have effectively evolved into full-on co-located conferences.)<br />
<br />
To me, the key difference is whether the workshop is structured around invited talks, panels, discussion (NIPS-style) or around contributed, reviewed submissions (ACL-style).<br />
<br />
For example, Paul Mineiro, Amanda Stent, Jason Weston and I are organizing a workshop at NIPS this year on ML for dialogue systems, <a href="http://letsdiscussnips2016.weebly.com/">Let's Discuss</a>. We have seven <i>amazing</i> <a href="http://letsdiscussnips2016.weebly.com/invited-speakers.html">invited speakers</a>: Marco Baroni, Antoine Bordes, Nina Dethlefs, Raquel Fernández, Milica Gasic, Helen Hastie and Jason Williams. If you look at our <a href="http://letsdiscussnips2016.weebly.com/schedule.html">schedule</a>, we have allocated 280 minutes to invited talks, 60 minutes to panel discussion, and 80 minutes to contributed papers. This is a quintessential NIPS-style workshop.<br />
<br />
For contrast, a standard ACL-style workshop might have one or two invited talks with the majority of the time spent on contributed (submitted/reviewed) papers.<br />
<br />
The difference in structure between NIPS-style and ACL-style workshops has some consequences:<br />
<ul>
<li>Reviewing in the NIPS-style tends to be very light, often without a PC, and often just by the organizers.</li>
<li>NIPS-style contributed workshop papers tend to be shorter.</li>
<li>NIPS-style workshop papers are almost always non-archival.</li>
</ul>
My personal experience is that NIPS-style workshops are a lot more fun. Contributed papers at ACL-style workshops are often those that might not cut it at the main conference. (Note: this isn't <i>always</i> the case, but it's common. It's also less often the case at workshops that represent topics that are not well represented at the main conference.) On the other hand, when you have seven invited speakers who are all experts in their field and were chosen by hand to represent a diversity of ideas, you get a much more intellectually invigorating experience.<br />
<br />
(Side note: my experience is that many many NIPS attendees only attend workshops and skip the main conferences; I've rarely heard of this happening at ACL. Yes, I could go get the statistics, but they'd be incomparable anyway because of time of year.)<br />
<br />
There are a few other effects that matter.<br />
<br />
The first is the archival-ness of ACL workshops which have proceedings that appear in the anthology:<br />
<div class="separator" style="clear: both; text-align: center;">
<a href="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEhAgS9kWzh98BK6ELJOeVm8V5rESh1VeizN1Qn-FGbKv5Rf2EiIfLMunlTPaLtF4fNEfrDqu48T87Y8zAtx3t7WSZevxG7u0RdY3AUxAweBNbJhJO3CqJQ29IWHvucAlyyOD15UNQ/s1600/Screenshot+from+2016-10-13+19-50-26.png" imageanchor="1" style="margin-left: 1em; margin-right: 1em;"><img border="0" height="186" src="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEhAgS9kWzh98BK6ELJOeVm8V5rESh1VeizN1Qn-FGbKv5Rf2EiIfLMunlTPaLtF4fNEfrDqu48T87Y8zAtx3t7WSZevxG7u0RdY3AUxAweBNbJhJO3CqJQ29IWHvucAlyyOD15UNQ/s400/Screenshot+from+2016-10-13+19-50-26.png" width="400" /></a></div>
(This is from EACL, but it's the same rules across the board.) I personally believe it's absurd that workshop papers are considered archival but papers on arxiv are not. By forcing workshop papers to be archival, you run the significant risk of guaranteeing that many submissions are things that authors have given up on getting into the main conference, which can lead to a weaker program.<br />
<br />
A second issue has to do with reviewing. Unfortunately as of about three years ago, the ACL organizing committee almost guaranteed that ACL workshops have to be ACL-style and not NIPS-style (personally I believe this is massive bureaucratic overreaching and micromanaging):<br />
<div class="separator" style="clear: both; text-align: center;">
<a href="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEgADSZcb3M9Q5ZcZWdFd9KNZJNB8cxKoPlZgrv0BpBJJu67SUhB9cJPXCGx1I9VVOekKzl-24xAz7VNazkYUuuLRlCF8AgH7F0wniL8fSKTSxyQJUP4gXgC-q6I41tICdsFGKCqmg/s1600/Screenshot+from+2016-10-15+13-30-41.png" imageanchor="1" style="margin-left: 1em; margin-right: 1em;"><img border="0" height="255" src="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEgADSZcb3M9Q5ZcZWdFd9KNZJNB8cxKoPlZgrv0BpBJJu67SUhB9cJPXCGx1I9VVOekKzl-24xAz7VNazkYUuuLRlCF8AgH7F0wniL8fSKTSxyQJUP4gXgC-q6I41tICdsFGKCqmg/s400/Screenshot+from+2016-10-15+13-30-41.png" width="400" /></a></div>
By forcing a program committee and reviewing, we're largely locked into the ACL-style workshop. Of course, some workshops ignore this and do more of a NIPS-style anyway, but IMO this should never have been a rule.<br />
<br />
One tricky issue with NIPS-style workshops is that, as I understand it, some students (and perhaps faculty/researchers) might be unable to secure travel funding to present at a non-archival workshop. I honestly have little idea how widespread this factor is, but if it's a big deal (e.g., perhaps in certain parts of the world) then it needs to be factored in as a cost.<br />
<br />
A second concern I have about NIPS-style workshops is making sure that they're inclusive. A significant failure mode is that of "I'll just invite my friends." In order to prevent this outcome, the workshop organizers have to make sure that they work hard to find invited speakers who are not necessarily in their narrow social networks. Having a broader set of workshop organizers can help. I think that when NIPS-style workshops are proposed, they should be required to list potential invited speakers (even if these people have not yet been contacted) and a significant part of the review process should be to make sure that these lists represent a diversity of ideas and a diversity of backgrounds. In the best case, this can lead to a <i>more</i> inclusive program than ACL-style workshops (where basically you get whatever you get as submissions) but in the worst case it can be pretty horrible. There are lots of examples of pretty horrible at NIPS in the past few years.<br />
<br />
At any rate, these aren't easy choices, but my preference is strongly for the NIPS-style workshop. At the very least, I don't think that ACL should predetermine which type is allowed at its conferences.<br />
<br />
<br />halhttp://www.blogger.com/profile/02162908373916390369noreply@blogger.com4tag:blogger.com,1999:blog-19803222.post-20764810436005227192016-11-08T17:06:00.002-07:002016-11-08T17:06:58.604-07:00Bias in ML, and Teaching AI<style type="text/css">p { margin-bottom: 0.1in; line-height: 120%; }a:link { }</style>
Yesterday I gave a super duper high level 12 minutes presentation
about some issues of bias in AI. I should emphasize (if it's not
clear) that this is something I am <i>not</i> an expert in; most of
what I know is by reading great papers by other people (there is a
completely non-academic sample at the end of this post). This blog
post is a variant of that presentation.<br />
<br />
Structure: most of the images below are prompts for talking
points, which are generally written below the corresponding image. I
think I managed to link all the images to the original source (let me
know if I missed one!).<br />
<br />
<span style="font-size: large;"><b>Automated Decision
Making is Part of Our Lives</b></span><br />
<br />
To me, AI is largely the study of <i>automated decision making</i>,
and the investment therein has been growing at a dramatic rate.
<br />
<br />
<div class="separator" style="clear: both; text-align: center;">
<a href="http://www.nature.com/news/there-is-a-blind-spot-in-ai-research-1.20805" imageanchor="1" style="margin-left: 1em; margin-right: 1em;"><img border="0" height="306" src="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEgZWYXcG4iYSMIOapjCb4Ro4PUTAINcqfr4iglzPqRN8cwTFihfGyymfv1ozmGl1wYIsaCbCYLQH5f_wKvei2GVMPkOXrt4zcojPAeUbwVp4qxcVEeP3KQgqRrkvH3nxc74JT9vWw/s400/on_the_rise.jpg" width="400" /></a></div>
<div style="text-align: right;">
<a href="http://www.nature.com/news/there-is-a-blind-spot-in-ai-research-1.20805">Source: Nature </a></div>
I'm currently teaching undergraduate artificial intelligence. The
last time I taught this class was in 2012. The amount that's changed
since there is incredible. Automated decision making is now a part of
basically everyone's life, and will only be more so over time. The
investment is in the billions of dollars per year.<br />
<br />
<span style="font-size: large;"><b>Things Can Go Really
Badly</b></span>
<br />
<br />
If you've been paying attention to headlines even just over the
past year, the number of high stakes settings in which automated
decisions are being made is growing, and growing into areas that
dramatically affect real people's real life, their well being, their
safety, and their rights.<br />
<div class="separator" style="clear: both; text-align: center;">
<a href="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEhL_UQqGN6xNUIw03kPUJ8ClTaEC6_hipW6lT6pIqOIFwDxHeQMbVHij45AXg9Z4HmdLChy-Se8dwws4kMVe8FvX9Ehb_uaZts9aezr5JObf0ZZjBm6rqMyTc1_M8CemKrMKACcGg/s1600/go_badly.png" imageanchor="1" style="margin-left: 1em; margin-right: 1em;"><img border="0" height="292" src="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEhL_UQqGN6xNUIw03kPUJ8ClTaEC6_hipW6lT6pIqOIFwDxHeQMbVHij45AXg9Z4HmdLChy-Se8dwws4kMVe8FvX9Ehb_uaZts9aezr5JObf0ZZjBm6rqMyTc1_M8CemKrMKACcGg/s400/go_badly.png" width="400" /></a></div>
<br />
This includes:<br />
<ul>
<li>
<div style="margin-bottom: 0in;">
<a href="http://onlinelibrary.wiley.com/doi/10.1111/j.1740-9713.2016.00960.x/full">Predictive
policing</a> in Oakland
</div>
</li>
<li>
<div style="margin-bottom: 0in;">
<a href="http://www.autoblog.com/2011/05/31/women-voice-command-systems/">Car
systems that are unable to understand women</a>
</div>
</li>
<li>
<div style="margin-bottom: 0in;">
<a href="https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2208240">Discriminatory
advertisement</a>s based on racially profiled name searches
</div>
</li>
<li>
<div style="margin-bottom: 0in;">
<a href="https://www.washingtonpost.com/news/wonk/wp/2016/03/10/uber-seems-to-offer-better-service-in-areas-with-more-white-people-that-raises-some-tough-questions/">Service
availability for transportation</a> (shout-out to <a href="http://towcenter.org/research/jennifer-stark/">Jennifer
Stark</a> & <a href="http://www.nickdiakopoulos.com/">Nick
Diakopolous</a> here at UMD!)
</div>
</li>
<li>
<div style="margin-bottom: 0in;">
<a href="https://www.propublica.org/article/machine-bias-risk-assessments-in-criminal-sentencing">Housing
discrimination</a> (which is straight up illegal)
</div>
</li>
<li>
<a href="https://www.propublica.org/article/machine-bias-risk-assessments-in-criminal-sentencing">Prediction
of criminality</a>
<br />
</li>
</ul>
This is obviously just a sample of some of the higher profile work
in this area, and while all of this is work in progress, even if
there's no impact today (hard to believe for me) it's hard to imagine
that this isn't going to be a major societal issue in the very near
future.<br />
<br />
<span style="font-size: large;"><b>Three (out of many)
Source of Bias</b></span><br />
<br />
For the remainder, I want to focus on three specific ways that
bias creeps in. The first I'll talk about more because we understand
it more, and it's closely related to work that I've done over the
past ten years or so, albeit in a different setting. These three are:<br />
<ol>
<li>
<div style="margin-bottom: 0in;">
data collection
</div>
</li>
<li>
<div style="margin-bottom: 0in;">
objective function
</div>
</li>
<li>
feedback loops<br />
<br />
</li>
</ol>
<span style="font-size: large;"><b>Sample Selection Bias</b></span><br />
<br />
The standard way that machine learning works is to take some
samples from a population you care about, run it through a machine
learning algorithm, to produce a predictor.<br />
<br />
<div class="separator" style="clear: both; text-align: center;">
<a href="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEhRvGEFG9qHqoenhsy7Ul2umnGOcGW3FejRnon3jWBmy2tqOwNYQn3BvBaDrnonMaWf-dw8ZViHpQdBFwS4aYqjvWrXmWJ7FcfcERCK_ruV1luQ9meFtI1MASfP5p8pgUE0GQ0krA/s1600/sel_pop.png" imageanchor="1" style="margin-left: 1em; margin-right: 1em;"><img border="0" height="105" src="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEhRvGEFG9qHqoenhsy7Ul2umnGOcGW3FejRnon3jWBmy2tqOwNYQn3BvBaDrnonMaWf-dw8ZViHpQdBFwS4aYqjvWrXmWJ7FcfcERCK_ruV1luQ9meFtI1MASfP5p8pgUE0GQ0krA/s400/sel_pop.png" width="400" /></a></div>
<br />
The magic of statistics is that if you then take new samples from
that same population, then, with high probability, the predictor will
do a good job. This is true for basically all models of machine
learning.<br />
<br />
The problem that arises is when your population samples are from a
subpopulation (or different population) for those on which you're
going to apply your predictor.<br />
<br />
<div class="separator" style="clear: both; text-align: center;">
<a href="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEiFlrHA9mqD7FL1i8hKp-3CFnFuf2BTszu_LPPmzPZezd4NX1Je9Isamwh0V_kb6bZyoIsx9kWZ98CNV4BJCWCt3GTKmQTwP2JbFPKjvTuDXl1aKMYnT3wflmei7obHeMxyTgaS2A/s1600/true_pop.png" imageanchor="1" style="margin-left: 1em; margin-right: 1em;"><img border="0" height="132" src="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEiFlrHA9mqD7FL1i8hKp-3CFnFuf2BTszu_LPPmzPZezd4NX1Je9Isamwh0V_kb6bZyoIsx9kWZ98CNV4BJCWCt3GTKmQTwP2JbFPKjvTuDXl1aKMYnT3wflmei7obHeMxyTgaS2A/s400/true_pop.png" width="400" /></a></div>
<br />
Both of my parents work in marketing research and have spent a lot
of their respective careers doing focuses groups and surveys. A few
years ago, my dad had a project working for a European company that
made skin care products. They wanted to break into the US market, and
hired him to conduct studies of what the US population is looking for
in skin care. He told them that he would need to conduct four or five
different studies to do this, which they gawked at. They wanted one
study, perhaps in the midwest (Cleveland or Chicago). The problem is
that skin care needs are very different in the southwest (moisturizer
matters) and the northwest (not so much), versus the northeast and
southeast. Doing one study in Chicago and hoping it would generalize
to Arizona and Georgia is unrealistic.<br />
<br />
This problem is often known as <i>sample selection bias</i> in the
statistics community. It also has other names, like covariate shift
and domain adaptation depending on who you talk to.<br />
<br />
<div class="separator" style="clear: both; text-align: center;">
<a href="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEir3zyYEYPwFyZPFNuqq-ZMbNvoKyM3eJ1l3k6Jx9rALdJUDF_YKaVmUY0P9vuk6TYu3V5-Dybw6IxW7liJ4597PVmDxSnMxHyMxd3vpCuSC4hqDN0_gwuBy99bAn5TjkwDutJRPA/s1600/heckman_cortes.png" imageanchor="1" style="margin-left: 1em; margin-right: 1em;"><img border="0" height="112" src="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEir3zyYEYPwFyZPFNuqq-ZMbNvoKyM3eJ1l3k6Jx9rALdJUDF_YKaVmUY0P9vuk6TYu3V5-Dybw6IxW7liJ4597PVmDxSnMxHyMxd3vpCuSC4hqDN0_gwuBy99bAn5TjkwDutJRPA/s400/heckman_cortes.png" width="400" /></a></div>
<br />
One of the most influential pieces of work in this area is the
1979 Econometrica paper by James Heckman, for which he won the 2000
Nobel Prize in economics. He's pretty happy about that! If you
haven't read <a href="https://www.jstor.org/stable/1912352?seq=1#page_scan_tab_contents">this
paper,</a> you should: it's only about 7 pages long, it's not that
difficult, and you won't believe the footnote in the last section.
(Sorry for the clickbait, but you really should read the paper.)<br />
There's been a ton of work in machine learning land over the past
twenty years, much of which builds on Heckman's original work. To
highlight one specific paper: Corinna Cortes is the Head of Google Research New York and has had
a number of excellent papers on this topic over the past ten years.
One in particular is <a href="http://static.googleusercontent.com/media/research.google.com/en//pubs/archive/44193.pdf">her
2013 paper</a> in Theoretical Computer Science (with Mohri) which
provides an amazingly in depth overview and new algorithms. Also a
highly recommended read.<br />
<br />
<span style="font-size: large;"><b>It's Not Just that Error
Rate Goes Up</b></span><br />
<br />
When you move from one sample space (like the southwest) to
another (like the northeast), you should first expect error rates to
go up.<br />
<br />
Because I wanted to run some experiments for this talk, here are
some simple adaptation numbers for predicting sentiment on Amazon
reviews (data due to <a href="https://www.cs.jhu.edu/%7Emdredze/datasets/sentiment/">Mark
Dredze and colleagues</a>). Here we have four domains (books, DVDs,
electronics and kitchen appliances) which you should think as
standins for the different regions of the US, or different
demographic qualifiers.<br />
<br />
<div class="separator" style="clear: both; text-align: center;">
</div>
<div class="separator" style="clear: both; text-align: center;">
<a href="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEjvQkHUyeFnJvLksK0vVRVcrJzfh5CCZWe0_7q3MxMYykqA57YCsStwphAfuVEHnVdmTwKacSxb7JoH7B2MnIszh0k58D4uDy3BRlwOGLiRHthwzuvOCGgIjMSqFpJu6Dw17XC0pA/s1600/cross_domain.png" imageanchor="1" style="margin-left: 1em; margin-right: 1em;"><img border="0" src="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEjvQkHUyeFnJvLksK0vVRVcrJzfh5CCZWe0_7q3MxMYykqA57YCsStwphAfuVEHnVdmTwKacSxb7JoH7B2MnIszh0k58D4uDy3BRlwOGLiRHthwzuvOCGgIjMSqFpJu6Dw17XC0pA/s1600/cross_domain.png" /></a></div>
<br />
The figure shows error rates when you train on one domain
(columns) and test on another (rows). The error rates are normalized
so that we have ones on the diagonal (actual error rates are about
10%). The off-diagonal shows how much additional error you suffer due
to sample selection bias. In particular, if you're making predictions
about kitchen appliances and <i>don't train on kitchen appliances</i>,
your error rate can be more than two times what it would have been.<br />
<br />
But that's not all.<br />
<br />
<div class="separator" style="clear: both; text-align: center;">
<a href="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEheddh5TGNUUk-40OpnSOec28u4BNzsSTWkXDcTafFv7GLXzVzIfZYDk6gpjV2BsoKVrT_SEeumS0SXCuqCUt8YOuIQAlQ1t9z3yn6jlNst9rtjqLLIvKRW5Djw_Dj7yHLT0e2Npw/s1600/electronics.png" imageanchor="1" style="margin-left: 1em; margin-right: 1em;"><img border="0" height="320" src="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEheddh5TGNUUk-40OpnSOec28u4BNzsSTWkXDcTafFv7GLXzVzIfZYDk6gpjV2BsoKVrT_SEeumS0SXCuqCUt8YOuIQAlQ1t9z3yn6jlNst9rtjqLLIvKRW5Djw_Dj7yHLT0e2Npw/s320/electronics.png" width="248" /></a></div>
<br />
These data sets are balanced: 50% positive, 50% negative. If you
train on electronics and make predictions on other domains, however,
you get different false positive/false negative rates. This shows the
number of test items predicted positively; you should expect it to be
50%, which basically is what happens in electronics and DVDs.
However, if you predict on books, you <i>underpredict</i> positives;
while if you predict on kitchen, you <i>overpredict</i> positives.<br />
<br />
So not only do the error rates go up, but the way they are
exhibited chances, too. This is closely related to issues of
disparate impact, which have been studied recently by many people,
for instance by <a href="https://arxiv.org/pdf/1412.3756.pdf">Feldman,
Friedler, Moeller, Scheidegger and Venkatasubramanian</a>.<br />
<br />
<span style="font-size: large;"><b>What Are We Optimizing
For</b></span><br />
<br />
One thing I've been trying to get undergrads in my AI class to
think about is what are we optimizing for, and whether the thing
that's being optimized for is what is best for us.<br />
<br />
<div class="separator" style="clear: both; text-align: center;">
<a href="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEgWYjaUPdbe26J1Fj3KdW7EivNR3AVCka8986wAKK4NcoN2ffdYHVMWsKHZJ_YTHejjc0Nvhd2RxnHZJrOieliEPQYDu3vd0U4pRMt3kdJQciDjX9tUD_Vwab17fz54cD58SS2mgA/s1600/map.png" imageanchor="1" style="margin-left: 1em; margin-right: 1em;"><img border="0" height="320" src="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEgWYjaUPdbe26J1Fj3KdW7EivNR3AVCka8986wAKK4NcoN2ffdYHVMWsKHZJ_YTHejjc0Nvhd2RxnHZJrOieliEPQYDu3vd0U4pRMt3kdJQciDjX9tUD_Vwab17fz54cD58SS2mgA/s320/map.png" width="287" /></a></div>
<br />
One of the first things you learn in a data structures class is
how to do graph search, using simple techniques like breadth first
search. In intro AI, you often learn more complex things like A*
search. A standard motivating example is how to find routes on a map,
like the planning shown above for me to drive from home to work
(which I never do because I don't have a car and it's slower than
metro anyway!).<br />
<br />
We spend a lot of time proving optimality of algorithms in terms
of shortest path costs, for fixed costs that have been given to us by
who-knows-where. I challenged my AI class to come up with features
that one might use to construct these costs. They started with
relatively obvious things: length of that segment of road, wait time
at lights, average speed along that road, whether the road is
one-way, etc. After more pushing, they came up with other ideas, like
how much gas mileage one gets on that road (either to save the
environment or to save money), whether the area is “dangerous”
(which itself is fraught with bias), what is the quality of the road
(ill-repaired, new, etc.).<br />
<br />
You can tell that my students are all quite well behaved. I then
asked them to be evil. Suppose you were an evil company, how might
you come up with path costs. Then you get things like: maybe
businesses have paid me to route more customers past their stores.
Maybe if you're driving the brand of car that my company owns or has
invested it, I route you along better (or worse) roads. Maybe I route
you so as to avoid billboards from competitors.<br />
<br />
The point is: we don't know, and there's no strong reason to a
priori assume that what the underlying system is optimizing for is my
personal best interest. (I should note that I'm definitely not saying
that Google or any other company is doing any of these things: just
that we should not assume that they're not.)<br />
<br />
<div class="separator" style="clear: both; text-align: center;">
<a href="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEhPBdbYKKEY5fu7sqSL2FVISPBJYg_hklsWJff4CnfNHk2eCe1RnyGkivp_nuH1j2uGlncRtoxKTkgA2xOHjEuIuzGDNUpuInqajsUPAzhHcuf2Ol9Pcq4hlOaRmMmGckRq68O2-w/s1600/crrrcuit.png" imageanchor="1" style="margin-left: 1em; margin-right: 1em;"><img border="0" height="320" src="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEhPBdbYKKEY5fu7sqSL2FVISPBJYg_hklsWJff4CnfNHk2eCe1RnyGkivp_nuH1j2uGlncRtoxKTkgA2xOHjEuIuzGDNUpuInqajsUPAzhHcuf2Ol9Pcq4hlOaRmMmGckRq68O2-w/s320/crrrcuit.png" width="231" /></a></div>
<br />
A more nuanced example is that of a dating application for, e.g.,
multi-colored robots. You can think of the color as representing any
sort of demographic information you like: political leaning (as
suggested by the red/blue choice here), sexual orientation, gender,
race, religion, etc. For simplicity, let's assume there are way more
blue robots than others, and let's assume that robots are at least
somewhat homophilous: they tend to associate with other similar
robots.<br />
<br />
If my objective function is something like “maximize number of
swipe rights,” then I'm going to want to disproportionately show
blue robots because, on average, this is going to increase my
objective function. This is especially true when I'm predicting
complex behaviors like robot attraction and love, and I don't have
nearly enough features to do anywhere near a perfect matching.
Because red robots, and robots of other colors, are more rare in my
data, my bottom line is not affected greatly by whether I do a good
job making predictions for them or not.<br />
<br />
<div class="separator" style="clear: both; text-align: center;">
<a href="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEhmFNtN2Vgmz_nZgUdfaeTEx8N9tAswL0ZsmjV83-p80TtTMEtD6Kb-_KqdD2HUC7SUBV7egye8S693DWN_HJleRnv_FHwlFc4SGx_3rsRrO8HTE0k65-lwoTMWcDqjeRvmlcOvTQ/s1600/vc.jpg" imageanchor="1" style="margin-left: 1em; margin-right: 1em;"><img border="0" src="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEhmFNtN2Vgmz_nZgUdfaeTEx8N9tAswL0ZsmjV83-p80TtTMEtD6Kb-_KqdD2HUC7SUBV7egye8S693DWN_HJleRnv_FHwlFc4SGx_3rsRrO8HTE0k65-lwoTMWcDqjeRvmlcOvTQ/s1600/vc.jpg" /></a><a href="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEgRiCTV6zBZsg3DyZqNLUMcJiwwCrfV209TwFQLYsjg3A8_6Mw1p20NVx7eZSJdyC-iGN5KC2mOYwtmg-lvvWoUmN6QiS2YUb2hO-BPcQ2i7DokxYriQdoz8JYf4PATU4g5XKTQHQ/s1600/palmer.jpg" imageanchor="1" style="margin-left: 1em; margin-right: 1em;"><img border="0" src="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEgRiCTV6zBZsg3DyZqNLUMcJiwwCrfV209TwFQLYsjg3A8_6Mw1p20NVx7eZSJdyC-iGN5KC2mOYwtmg-lvvWoUmN6QiS2YUb2hO-BPcQ2i7DokxYriQdoz8JYf4PATU4g5XKTQHQ/s1600/palmer.jpg" /></a></div>
<br />
I highly recommend reading <a href="http://dexterpalmer.com/books/#version">Version
Control</a>, a recent novel by Dexter Palmer. I especially recommend
it if you have, or will, teach AI. It's fantastic.<br />
There is an interesting vignette that Palmer describes (don't
worry, no plot spoilers) in which a couple engineers build a dating
service, like Crrrcuit, but for people. In this thought exercise, the
system's objective function is to help people find true love, and
they are wildly successful. They get investors. The investors realize
that when their product succeeds, they lose business. This leads to a
more nuanced objective in which you want to match <i>most</i><span style="font-style: normal;">
people (to maintain trust), but not perfectly (to maintain
clientèle). But then, to make money, the company starts selling its
data to advertisers. And different individuals' data may be more
valuable: in particular, advertisers might be willing to pay a lot
for data from members of underrepresented groups. This provides
incentive to actively do a </span><i>worse</i><span style="font-style: normal;">
job than usual on such clients. In the book, this thought exercise
proceeds by human reasoning, but it's pretty easy to see that if one
set up, say, a reinforcement learning algorithm for predicting
matches that had long term company profit as its objective function,
it could learn something similar and we'd have no idea that that's
what the system was doing.</span><br />
<br />
<span style="font-size: large;"><b>Feedback Loops</b></span><br />
<br />
<a href="http://www.rshroff.com/"><span style="font-style: normal;">R</span></a><span style="font-style: normal;"><a href="http://www.rshroff.com/">avi Shroff</a> recently visited the CLIP lab and talked about his work
(with Justin Rao and Shared Goel) related to <a href="http://www.rshroff.com/uploads/6/2/3/5/62359383/frisky.pdf">stop and frisk policies in New York</a>. The setup here is that the “stop and frisk”
rule (in 2011, <a href="https://en.wikipedia.org/wiki/Stop-and-frisk_in_New_York_City">over 685k people were stopped</a>; this has subsequently been declared unconstitutional in New York) gave police
officers the right to stop people with much lower thresholds than probable cause, to try
to find contraband weapons or drugs. Shroff and colleagues focused on
weapons.</span><br />
<br />
<div style="font-style: normal;">
They considered the following model: a
police officer sees someone behaving strangely, and decide that they
want to stop and frisk that person. Before doing so, they enter a few
values into their computer, and the computer either gives a thumbs up
(go ahead and stop) or a thumbs down (let them live their life). One
question was: can we cut down on the number of stops (good for
individuals) while still finding most contraband weapons (good for
society)?<br />
</div>
<div class="separator" style="clear: both; text-align: center;">
<a href="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEiNY8t0UiQ3AmvQ3LrSQ8xBDl0b4QzqxjKldVTucbDFI_zbDos87cR0lQepNFp-6mSvKvTNxWXljMkfa_SdJo-5-hkgOdNOVcejRbvRU7pRQxkUz62AFqntz4vmxy5ovxLjJDEsjA/s1600/shroff.png" imageanchor="1" style="margin-left: 1em; margin-right: 1em;"><img border="0" height="300" src="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEiNY8t0UiQ3AmvQ3LrSQ8xBDl0b4QzqxjKldVTucbDFI_zbDos87cR0lQepNFp-6mSvKvTNxWXljMkfa_SdJo-5-hkgOdNOVcejRbvRU7pRQxkUz62AFqntz4vmxy5ovxLjJDEsjA/s320/shroff.png" width="320" /></a></div>
<br />
<span style="font-style: normal;">In this figure, we can see that
if the system thumbs downed 90% of stops (and therefore only 10% of
people that police </span><i>would have</i><span style="font-style: normal;">
stopped get stopped), they are still able to recover about 50% of the
weapons. </span><span style="font-style: normal;">With stopping only
about 1/3 of individuals, they are able to recover 75% of weapons.
This is a </span><i>massive</i><span style="font-style: normal;">
reduction in privacy violations while still successfully keeping the
majority of weapons off the streets.</span><br />
<br />
<i>(Side note: you might worry about sample selection bias here,
because the models are trained on people that the policy did actually
stop. Shroff and colleagues get around this by the assumption I
stated before: the model is only run on people who policy have
already decided are suspicious and would have stopped and frisked
anyway.)</i><br />
<br />
The question is: what happens if and when such a system is
deployed in practice?<br />
<br />
The issue is that policy officers, like humans in general, are not
stationary entities. Their behavior changes over time, and it's
reasonable to assume that their behavior would change when they get
this new system. They might feed more people into the system (in
“hopes” of thumbs up) or feed fewer people into the system
(having learned that the system is going to thumbs down them anyway).
This is similar to how the sorts of queries people issue against web
search engines change over time, partially because we learn to use
the systems more effectively, and learn what to not consider asking a
search engine to do for us because we know it will fail.<br />
<br />
Now, once we've (hypothetically) deployed this system, it's
collecting its own data, which is going to be fundamentally different
from the data is was originally trained one. It can continually
adapt, but we need good technology for doing this that takes into
account the human behavior of the officers.<br />
<br />
<span style="font-size: large;"><b>Wrap Up and Discussion</b></span><br />
<div style="font-style: normal;">
<br />
There are many things that I didn't
touch on above that I think are nonetheless really important. Some
examples:</div>
<ol>
<li>
<div style="font-style: normal;">
All the example “failure”
cases I showed above have to do with race or (binary) gender. There
are other things to consider, like sexual orientation, religion,
political views, disabilities, family and child status, first
language, etc. I tried and failed to find examples of such things,
and would appreciate pointers. For instance, I can easily imagine
that speech recognition error rates skyrocket when working for users
with speech impairments, or with non-standard accents, or who speak
a dialect of English that not the “status quo academic English.”
I can also imagine that visual tracking of people might fail badly
on people with motor impairments or who use a wheelchair.<br /></div>
</li>
<li>
<div style="font-style: normal;">
I am particularly concerned
about less “visible” issues because we might not even know. The
standard example here is: could a social media platform sway an
election by reminding people who (it believes) belong to a
particular political party to vote? How would we even know?<br /></div>
</li>
<li>
<div style="font-style: normal;">
We need to start thinking about
qualifying our research better with respect to the populations we
expect it to work on. When we pick a problem to work on, who is
being served? When we pick a dataset to work on, who is being left
out? A silly example is the curation of older datasets for object
detection in computer vision, which (I understand) decided on which
objects to focus on by asking five year old relatives of the
researchers constructing the datasets to name all the objects they
could see. As a result of socio-economic status (among other
things), mouse means the thing that attaches to your computer, not
the cute furry animal. More generally, when we say we've “solved”
task X, does this really mean task X or does this mean task X for
some specific population that we haven't even thought to identify
(i.e., “people like me” aka the <a href="http://www.nytimes.com/2016/06/26/opinion/sunday/artificial-intelligences-white-guy-problem.html">white
guys problem</a>)? And does “getting more data” really solve the
problem---is more data always good data?<br /></div>
</li>
<li>
<div style="font-style: normal;">
I'm at least as concerned with
machine-in-the-loop decision making as fully automated decision
making. Just because a human makes the final decision doesn't mean
that the system cannot bias that human. For complex decisions, a
system (think even just web search!) has to provide you with
information that helps you decide, but what guarantees do we have
that that information isn't going to be biased, either
unintentionally or even intentionally. (I've also heard that, e.g.,
in predicting recidivism, machine-in-the-loop predictions are worse
than fully automated decisions, presumably because of some human
bias we don't understand.)</div>
</li>
</ol>
<div style="font-style: normal;">
If you've read this far, I hope you've
found some things to think about. If you want more to read, here are
some people whose work I like, who tweet about these topics, and for
whom you can citation chase to find other cool work. It's a highly
biased list.</div>
<ul>
<li>
<div style="font-style: normal;">
<a href="http://www.cs.bath.ac.uk/~jjb/">Joanna Bryson</a> (<a href="http://twitter.com/j2bryson">@j2bryson</a>), who
has been doing great work in ethics/AI for a long time and whose
work on bias in language has given me tons of food for thought.</div>
</li>
<li>
<div style="font-style: normal;">
<a href="http://www.katecrawford.net/">Kate Crawford</a> (<a href="http://twitter.com/katecrawford">@katecrawford</a>)
studies the intersection between society and data, and has written
excellent pieces on fairness.</div>
</li>
<li>
<div style="font-style: normal;">
<a href="http://www.nickdiakopoulos.com/">Nick Diakopoulos</a>
(<a href="http://twitter.com/ndiakopoulos">@ndiakopoulos</a>), a colleague here at UMD, studies computational
journalism and algorithmic transparency.</div>
</li>
<li>
<div style="font-style: normal;">
<a href="http://sorelle.friedler.net/">Sorelle Friedler</a> (<a href="http://twitter.com/kdphd">@kdphd</a>), a
former PhD student here at UMD!, has done some of the initial work
on learning without disparate impact.</div>
</li>
<li>
<div style="font-style: normal;">
<a href="http://www.cs.utah.edu/~suresh">Suresh Venkatasubramanian</a>
(<a href="http://twitter.com/geomblog">@geomblog</a>) has co-authored many of the papers with Friedler,
including work on lower bounds and impossibility results for
fairness.</div>
</li>
<li>
<div style="font-style: normal;">
<a href="http://dirichlet.net/">Hanna Wallach</a> (<a href="http://twitter.com/hannawallach">@hannawallach</a>) is
the first name I think of for machine learning and computational
social science, and has recently been working in the area of
fairness.</div>
</li>
</ul>
<div style="font-style: normal;">
I'll also point to less biased sources.
The <a href="http://fatml.org/">Fairness, Accountability and
Transparency in Machine Learning workshop</a> takes place in New York
City in a week and a half; check out the speakers and papers there. I
also highly recommend the very long reading list on <a href="http://socialmediacollective.org/reading-lists/critical-algorithm-studies">Critical
Algorithm Studies</a>, which covers more than just machine learning.</div>
<div style="font-style: normal;">
<br />
<br /></div>
halhttp://www.blogger.com/profile/02162908373916390369noreply@blogger.com4tag:blogger.com,1999:blog-19803222.post-84342693089343885112016-11-02T07:12:00.001-06:002016-11-02T07:12:59.730-06:00ACL 2017 PC Chairs BlogIn case you missed it, Regina Barzilay and Min-Yen Kan, who are program chairs for ACL 2017, are running an open blog describing what they're doing:<br />
<br />
https://chairs-blog.acl2017.org/<br />
<br />
Check it out!halhttp://www.blogger.com/profile/02162908373916390369noreply@blogger.com1tag:blogger.com,1999:blog-19803222.post-11790444170381384822016-08-24T11:15:00.001-06:002016-08-24T11:22:56.156-06:00Debugging machine learningI've been thinking, mostly in the context of teaching, about how to specifically teach debugging of machine learning. Personally I find it very helpful to break things down in terms of the usual error terms: Bayes error (how much error is there in the best possible classifier), approximation error (how much do you pay for restricting to some hypothesis class), estimation error (how much do you pay because you only have finite samples), optimization error (how much do you pay because you didn't find a global optimum to your optimization problem). I've generally found that trying to isolate errors to one of these pieces, and then debugging that piece in particular (eg., pick a better optimizer versus pick a better hypothesis class) has been useful.<br />
<br />
For instance, my general debugging strategy involves steps like the following:<br />
<ol>
<li>First, ensure that your optimizer isn't the problem. You can do this by adding "cheating" features -- a feature that correlates perfectly with the label. Make sure you can successfully overfit the training data. If not, this is probably either an optimizer problem or a too-small-sample problem.</li>
<li>Remove all the features except the cheating feature and make sure you can overfit then. Assuming that works, add feature back in incrementally (usually at an exponential rate). If at some point, things stop working, then probably you have too many features or too little data.</li>
<li>Remove the cheating features and make your hypothesis class much bigger; e.g., by adding lots of quadratic features. Make sure you can overfit. If you can't overfit, maybe you need a better hypothesis class.</li>
<li>Cut the amount of training data in half. We usually see test accuracy asymptote as the training data size increases, so if cutting the training data in half has a huge effect, you're not yet asymptoted and you might do better to get some more data.</li>
</ol>
The problem is that this normal breakdown of error terms comes from theory land, and, well, sometimes theory misses out on some stuff because of a particular abstraction that has been taken. Typically this abstraction has to do with the fact that the overall goal has already been broken down into an iid/PAC style learning problem, and so you end up unable to see some types of error because the abstraction hides them.<br />
<br />
In an effort to try to understand this better, I tried to make a flow chart of sorts that encompasses all the various types of error I could think of that can sneak into a machine learning system. This is shown below:<br />
<div class="separator" style="clear: both; text-align: center;">
<a href="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEiMFX1LEzZXsVugi08PYsUWXK9PmMRHy6m6FZ7VDxKwblLp-zp-rMDUCw5YLsKgBpeZkTDwklKboVyPWsKMPoi9Cfah8T4kcwke_qNDjsiuwm94lyVyqmFddmWPzqxRHT9faJIYcg/s1600/deployml.png" imageanchor="1" style="margin-left: 1em; margin-right: 1em;"><img alt="Real world goal (increase revenue) -> 1. Real world mechanism (better ad display) -> 2. Learning problem (classify clickthrough) -> 3. Data colleciton mechanism (interactions with current system) -> 4. Collected data (query, ad, click) -> 5. Data representation (bag of words, click) -> 6. Hypothesis class/inductive bias choice (decision trees, depth 20) -> 7. Training data selection (subset from April 2016) -> 8. Model training + hyperparams (final decision tree) -> 9. Predict on test data (subset from May 2016) -> 10. Evaluate error (AUC for +- click prediction) -> 11. Deploy!" border="0" height="353" src="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEiMFX1LEzZXsVugi08PYsUWXK9PmMRHy6m6FZ7VDxKwblLp-zp-rMDUCw5YLsKgBpeZkTDwklKboVyPWsKMPoi9Cfah8T4kcwke_qNDjsiuwm94lyVyqmFddmWPzqxRHT9faJIYcg/s400/deployml.png" title="" width="400" /> </a></div>
<div class="separator" style="clear: both; text-align: left;">
I've tried to give some reasonable names to the steps (the left part of the box) and then give a grounded example in the context of ad placement (because it's easy to think about). I'll walk through the steps (1-11) and try to say something about what sort of error can arise at that step.</div>
<ol>
<li> In the first step, we take our real world goal of increasing revenue for our company and decide to solve it by improving our ad displays. This immediately upper bounds how much increased revenue we can hope for because, well, maybe ads are the wrong thing to target. Maybe I would do better by building a better product. This is sort of a "business" decision, but it's perhaps the most important question you can ask: am I even going after the right things?</li>
<li>Once you have a real world mechanism (better ad placement) you need to turn it into a learning problem (or not). In this case, we've decided that the way we're going to do this is by trying to predict clickthrough, and then use those predictions to place better ads. Is clickthrough a good thing to use to predict increased revenue? This itself is an active research area. But once you decide that you're going to predict clickthrough, you suffer some loss because of a mismatch between that prediction task and the goal of better ad placement.</li>
<li>Now you have to collect some data. You might do this by logging interactions with a currently deployed system. This introduces all sorts of biases because the data you're collecting is not from the final system you want to deploy (the one you're building now), and you will pay for this in terms of distribution drift.</li>
<li>You cannot possibly log everything that the current system is doing, so you have to only log a subset of things. Perhaps you log queries, ads, and clicks. This now hides any information that you didn't log, for instance time of day or day of week might be relevant, user information might be relevant, etc. Again, this upper bounds your best possible revenue.</li>
<li>You then usually pick a data representation, for instance quadratic terms between a bag of words on the query side and a bag of words on the ad side, paired with a +/- on whether the user clicked or not. We're now getting into the position where we can start using theory words, but this is basically limited the best possible Bayes error. If you included more information, or represented it better, you might be able to get a lower Bayes error.</li>
<li>You also have to choose a hypothesis class. I might choose decision trees. This is where my approximation error comes from.</li>
<li>We have to pick some training data. The real world is basically never i.i.d., so any data we select is going to have some bias. It might not be identically distributed with the test data (because things change month to month, for instance). It might not be independent (because things don't change much second to second). You will pay for this.</li>
<li>You now train your model on this data, probably tuning hyperparameters too. This is your usual estimation error.</li>
<li>We now pick some test data on which to measure performance. Of course, this test data is only going to be representative of how well your system will do in the future if this data is so representative. In practice, it won't be, typically at least because of concept drift over time.</li>
<li>After we make predictions on this test data, we have to choose some method for evaluating success. We might use accuracy, f measure, area under the ROC curve, etc. The degree to which these measures correlate with what we really care about (ad revenue) is going to affect how well we're able to capture the overall task. If the measure anti-correlates, for instance, we'll head downhill rather than uphill.</li>
</ol>
(Minor note: although I put these in a specific order, that's not a prescriptive order, and many can be swapped. Also, of course there are lots of cycles and dependencies here as one continues to improve systems.)<br />
<br />
Some of these things are active research areas. Things like sample selection bias/domain adaptation/covariate shift have to do with mismatch of train/test data. For instance, if I can overfit train but generalization is horrible, I'll often randomly shuffle train/test into a new split and see if generalization is better. If it is, there's probably an adaptation problem.<br />
<br />
When people develop new evaluation metrics (like Bleu for machine translation), they try to look at things like #10 (correlation with some goal, perhaps not exactly the end goal). And standard theory and debugging (per above) covers some of this too. <br />
<br />
I'm very curious if y'all have topics/tricks that you like that aren't mentioned here.<br />
<br />
Related reading:<br />
<ul>
<li><a href="https://www.opendatascience.com/blog/the-7-steps-of-a-data-project/">The Seven Steps of a Data Project</a>; Alivia Smith</li>
<li><a href="http://static.googleusercontent.com/media/research.google.com/en//pubs/archive/43146.pdf">Machine Learning: The High Interest Credit Card of Technical Debt</a>; Sculley et al.</li>
</ul>
halhttp://www.blogger.com/profile/02162908373916390369noreply@blogger.com1tag:blogger.com,1999:blog-19803222.post-15043506582591320962016-08-16T17:22:00.002-06:002016-08-16T17:22:19.710-06:00Feature (or architecture) ablationI wrote my first (and only) <a href="http://hal3.name/docs/daume05coref.pdf">coreference paper</a> back in 2005. At the time, my goals were to (a) do well on coref, (b) integrate background knowledge (like "Bush" is "president") using simple techniques, and (c) try to figure out how important different (types of) features were for making coreference decisions.<br />
<br />
For the last, there is a reasonably extensive feature-type ablation experiment using backward selection (which I trust far more than forward selection). After writing the paper, I had many internal dialogues about why experiments like that are interesting. I have had, over the years, a couple of answers:<br />
<ol>
<li>The obvious answer is "it tells us something interesting about language." It would be nice if this were true, but I'm not totally sure it is, and it's definitely not true if one doesn't put a bunch more effort into it than I put into that 2005 paper. What can we say? Yeah, spelling is important. Knowledge is important. Syntax is hard to actualize. I don't know that we didn't already know these things before.<br /></li>
<li>Engineering. Suppose someone wanted to build a similar system. They want to put their effort where it's most valuable, and so feature ablation experiments tell you where you're likely to get the most bang for the buck. In a sense, you can see these as<i> a type of negative result</i>. Which features actually aren't that important. In the 2005 paper, you could remove syntactic, semantic, and class-based features with zero performance degradation; and also get rid of pattern-based features with minor performance degradation. This saves a lot of effort because some of these are actually quite a pain to implement and/or are slow and/or require lots of external resources.</li>
</ol>
Today, I mostly lean toward the engineering answer, or at least that's what I want to use as a jumping off point here.<br />
<br />
Now that we're partially allergic to feature engineering and prefer to replace it with architecture engineering, I think the charge is stronger, not weaker, to do ablation experiments. Does that thing really need to be a biLSTM? Would an RNN suffice? What about just averaged bag of word embeddings? Do you need two layers of attention there or would one suffice? Do you need attention at all? Does that layer really need to be that wide?<br />
<br />
These are all easy questions to ablate and answer.<br />
<br />
There's never going to be a crisp answer like "yes, if I cut my hidden state from 493 units to 492 units performance goes down the drain." Many things will be gradual, but not all.<br />
<br />
Why do I think this is important? Precisely for reason #2 above, but about a bajillion times more so. Training these really complicated models with wide hidden units, bidirectional stuff, etc., is really slow. Really really slow. If you tell me I can be within 1% accuracy but can train 100 times faster, I'm going to do it. Sure, for a final test run I might crank everything up again (and then report that!) but for development, it's super useful to have a system you can train and evaluate efficiently.<br />
<br />
Does this tell us anything interesting about language? Almost certainly not (or at least not without a huge amount of extra work). But it does make everyone's life better.<br />
<br />halhttp://www.blogger.com/profile/02162908373916390369noreply@blogger.com4tag:blogger.com,1999:blog-19803222.post-71864813626055204382016-08-14T11:36:00.000-06:002016-08-14T12:42:22.083-06:00Some papers I liked at ACL 2016A conference just ended, so it's that time of year! Here are some papers I liked with the usual caveats about recall.<br />
<br />
Before I go to the list, let me say that I really really enjoyed ACL this year. I was completely on the fence about going, and basically decided to go only because of giving a talk at <a href="https://sites.google.com/site/repl4nlp2016/">Repl4NLP</a>, and wanted to attend the business meeting for the discussion of diversity in the ACL community, led by Joakim Nivre with an <i>amazing</i> report that he, Lyn Walker, Yejin Choi and Min-Yen Kan put together. (Likely I'll post more, separately, about both of these; for the latter, I tried to <a href="https://twitter.com/haldaume3/status/763351580821291008">transcribe much of Joakim's presentation</a>.)<br />
<br />
All in all, I'm supremely glad I decided to go: it was probably my favorite conference in recent memory. This was not just because there were lots of great papers (there were!) but also because somehow it felt more like a large community conference than others I've attended recently. I'm not sure what made it like this, but I noticed it felt a lot less clique-y than NAACL, a lot more broad and interesting than ICML/NIPS (though that's probably because of my personal taste in research) and in general a lot friendlier. I don't know what the organizers did that managed this great combination, but it was great!<br />
<br />
Okay, so on to the papers, sorted by ACL id.<br />
<br />
<a href="http://www.aclweb.org/anthology/P/P16/P16-1009.pdf">P16-1009</a>: <b>Rico Sennrich; Barry Haddow; Alexandra Birch</b><br />
<i>Improving Neural Machine Translation Models with Monolingual Data</i><br />
<blockquote class="tr_bq">
I like this paper because it has a nice solution to a problem I spent a year thinking about on-and-off and never came up with. The problem is: suppose that you're training a discriminative MT system (they're doing neural; that's essentially irrelevant). You usually have far more monolingual data than parallel data, which typically gets thrown away in neural systems because we have no idea how to incorporate it (other than as a feature, but that's blech). What they do here is, assuming you have translation systems in both directions, back translate your monolingual target-side data, and then use that faux-parallel-data to train your MT system on. Obvious question is: how much of the improvement in performance is due to language modeling versus due to some weird kind of reverse-self-training, but regardless the answer, this is a really cool (if somewhat computationally expensive) answer to a question that's been around for at least five years. Oh and it also works <i>really</i> well.<br />
<i></i></blockquote>
<br />
<a href="http://www.aclweb.org/anthology/P/P16/P16-1018.pdf">P16-1018</a>: <b>E.Dario Gutierrez; Ekaterina Shutova; Tyler Marghetis; Benjamin Bergen</b><br />
<i>Literal and Metaphorical Senses in Compositional Distributional Semantic Models</i>
<br />
<blockquote class="tr_bq">
I didn't see this paper presented, but it was suggested to me at Monday's poster session. Suppose we're trying to learn representations of adjective/noun pairs, by modeling nouns as vectors and adjectives as matrices, evaluating on unseen pairs only. (Personally I don't love this style, but that's incidental to the main ideas in this paper.) This paper adjusts the adjective matrices depending on whether they're being used literally ("sweet candy") or metaphorically ("sweet dreams"). But then you can go further and posit that there's another matrix that can transform literal metaphors into metaphorical metaphors automatically, essentially implementing the Lakoff-style notion that there is great consistency in how metaphors are created.</blockquote>
<br />
<a href="http://www.aclweb.org/anthology/P/P16/P16-1030.pdf">P16-1030</a> [<a href="http://www.aclweb.org/anthology/attachments/P/P16/P16-1030.Datasets.zip">dataset</a>]: <b>Hannah Rashkin; Sameer Singh; Yejin Choi</b><br />
<i>Connotation Frames: A Data-Driven Investigation</i><br />
<blockquote class="tr_bq">
This paper should win some sort of award for thoroughness. The idea is that in many frames ("The walrus pummelled the sea squirt") there is implied connotation/polarity/etc. on not only the agent (walrus) and theme (sea squirt) of the frame but also tells us something about the relationship between the writer/speaker and the agent/theme (the writer might be closer to the sea squirt in this example, versus s/pummelled/fought/). The connotation frame for pummelled collects all this information. This paper also describes an approach to prediction of these complex frames using nice structured models. Totally want to try this stuff on our old <a href="http://hal3.name/docs/daume10plotunits-emnlp.pdf">plotunits</a> data, where we had a hard time getting even a much simpler type of representation (patient polarity verbs) to work!</blockquote>
<br />
<a href="http://www.aclweb.org/anthology/P/P16/P16-1152.pdf">P16-1152</a>: <b>Artem Sokolov; Julia Kreutzer; Christopher Lo; Stefan Riezler</b><br />
<i>Learning Structured Predictors from Bandit Feedback for Interactive NLP</i><br />
<blockquote class="tr_bq">
This was perhaps my favorite paper of the conference because it's trying to do something new and hard and takes a nice approach. At a high level, suppose you're Facebook and you're trying to improve your translation system so you ask users to give 1 star to 5 star ratings. How can you use this to do better translation? This is basically the (structured) contextual bandit feedback learning problem. This paper approaches this from a dueling bandits perspective where I show you two translations and ask which is better. (Some of the authors had an earlier <a href="http://www.cl.uni-heidelberg.de/~riezler/publications/papers/MTSUMMIT2015.pdf">MT-Summit paper</a> on the non-dueling problem which I imagine many people didn't see, but you should read it anyway.) The technical approach is basically probabilitic latent-variable models, optimized with gradient descent, with promising results. (I also like this because I've been thinking about similar <a href="https://arxiv.org/abs/1502.02206">structured bandit problems</a> recently too :P.)</blockquote>
<br />
<a href="http://www.aclweb.org/anthology/P/P16/P16-1231.pdf">P16-1231</a>: <b>Daniel Andor; Chris Alberti; David Weiss; Aliaksei Severyn; Alessandro Presta; Kuzman Ganchev; Slav Petrov; Michael Collins</b><br />
<i>Globally Normalized Transition-Based Neural Networks</i>
<br />
<blockquote class="tr_bq">
[EDIT 14 Aug 2:40p: I misunderstood from the talk and therefore the following is basically inaccurate. I'm leaving this description and paper here on the list because Yoav's comment will make no sense otherwise, but please understand that it's wrong and, I hate to say this, it does make the paper less exciting to me. The part that's wrong is struck-out below.] There's a theme in the past two years of basically repeating all the structured prediction stuff we did ten years ago on our new neural network technology. This paper is about using Collins & Roark-style incremental perceptron for transition-based dependency parsing on top of neural networks. The idea is that label-bias is perhaps still a problem for neural network dependency parsers, like their linear grandparents. <strike>Why do I like this? Because I think a lot of neural nets people would argue that this shouldn't be necessary: the network can do arbitrarily far lookahead into the future and therefore should be able to learn to avoid the label-bias problem. This paper shows that current techniques don't achieve that: there's a consistent win to be had by doing global normalization.</strike></blockquote>
<br />
<a href="http://www.aclweb.org/anthology/P/P16/P16-2013.pdf">P16-2013</a>: <b>Marina Fomicheva; Lucia Specia</b><br />
<i>Reference Bias in Monolingual Machine Translation Evaluation</i><br />
<blockquote class="tr_bq">
This paper shows pretty definitively that human evaluations against a reference translation are super biased toward the particular reference used (probably because evaluators are lazy and are basically doing ngram matching anyway -- a story I heard from MSR friends a while back). The paper also shows that this gets worse over time, presumably as evaluators get tireder.</blockquote>
<a href="http://www.aclweb.org/anthology/P/P16/P16-2096.pdf">P16-2096</a>: <b>Dirk Hovy; Shannon L. Spruit</b><br />
<i>The Social Impact of Natural Language Processing</i><br />
<blockquote class="tr_bq">
This is a nice paper summarizing four issues that come up in ethics that also come up in NLP. I mostly liked this paper because it gave names to things I've thought about off and on, but didn't have a name for. In particular, they consider <i>exclusion</i> (hey my ASR system doesn't work on people with an accent, I guess they don't get a voice), <i>overgeneralization</i> (to what degree are our models effectively stereotyping more than they should), <i>over-</i> and <i>under-exposure</i> (hey lets all work on parsing because that's what everyone else is working on, which then makes parsing seem more important...just to pick a random example :P), and <i>dual-use</i> (I made something for good but XYZ organization used it for evil!). This is a position/discussion-starting paper, and I thought quite engaging.</blockquote>
<br />
Feel free to post papers you like in the comments! <br />
<br />
<br />
<br />
<br />
<br />
<i> </i> halhttp://www.blogger.com/profile/02162908373916390369noreply@blogger.com13tag:blogger.com,1999:blog-19803222.post-90644143533279964412016-08-05T10:27:00.004-06:002016-08-05T12:29:01.081-06:00Fast & easy baseline text categorization with vwAbout a month ago, the paper <a href="https://arxiv.org/pdf/1607.01759v2.pdf">Bag of Tricks for Efficient Text Categorization</a> was posted to arxiv. I found it thanks to <a href="https://twitter.com/yoavgo/status/751178795323908096">Yoav Goldberg's rather incisive tweet</a>:<br />
<br />
<div class="separator" style="clear: both; text-align: center;">
<a href="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEhIOq13mhE_YDAXBeMw8M4ZAghqTMhmun44TY64c7gDd9oQC5TszxIgEqQW-Bqi_7AVnWFHS8WBpo-DN3coM2u3QRV07isHalkPNj70YlNxepaw7OqD2f8UXsih9Fi0aAvhCIR8wQ/s1600/yaov.png" imageanchor="1" style="margin-left: 1em; margin-right: 1em;"><img border="0" height="110" src="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEhIOq13mhE_YDAXBeMw8M4ZAghqTMhmun44TY64c7gDd9oQC5TszxIgEqQW-Bqi_7AVnWFHS8WBpo-DN3coM2u3QRV07isHalkPNj70YlNxepaw7OqD2f8UXsih9Fi0aAvhCIR8wQ/s400/yaov.png" width="400" /></a></div>
Yoav is basically referring to the fact that the paper is all about (a) hashing features and (b) bigrams and (c) a projection that doesn't totally make sense to me, which (a) vw does by default (b) requires "--ngrams 2" and (c) I don't totally understand I don't think is necessary. (See <a href="https://github.com/hal3/vwnlp">this tutorial</a> for more on how to do NLP in VW.)<br />
<br />
At the time, I said if they gave me the data, I'd run <a href="https://github.com/JohnLangford/vowpal_wabbit">vw</a> on it and report results. They were nice enough to share the data but I never got around to running it. The code for their technique ("fastText") was just released, which goaded me into finally doing something.<br />
<br />
So my goal here was to try to tell, without tuning any parameters, how competitive a baseline vw is to the results from fastText with minimal effort.<br />
<br />
Here are the results:<br />
<br />
<table border="1" cellpadding="2" cellspacing="0">
<tbody>
<tr><td></td><td></td><td colspan="2" style="text-align: center;"><b>fastText</b></td><td colspan="2" style="text-align: center;"><b>vw</b></td></tr>
<tr><td style="text-align: center;"><b>Dataset</b></td><td style="text-align: center;"><b>ng</b></td><td style="text-align: center;"><b>time</b></td><td style="text-align: center;"><b>acc</b></td><td style="text-align: center;"><b>time</b></td><td style="text-align: center;"><b>acc</b></td></tr>
<tr><td>ag news</td><td style="text-align: center;">1</td><td style="text-align: right;"><br /></td><td style="text-align: right;">91.5</td><td>2s</td><td style="text-align: right;"><b>91.9</b></td></tr>
<tr><td>ag news</td><td style="text-align: center;">2</td><td style="text-align: right;">3s</td><td style="text-align: right;"><b>92.5</b></td><td>5s</td><td style="text-align: right;">92.3</td></tr>
<tr><td>amazon full</td><td style="text-align: center;">1</td><td style="text-align: right;"><br /></td><td style="text-align: right;"><b>55.8</b></td><td>47s</td><td style="text-align: right;">53.6</td></tr>
<tr><td>amazon full</td><td style="text-align: center;">2</td><td style="text-align: right;">33s</td><td style="text-align: right;"><b>60.2</b></td><td>69s</td><td style="text-align: right;">56.6</td></tr>
<tr><td>amazon polarity</td><td style="text-align: center;">1</td><td style="text-align: right;"><br /></td><td style="text-align: right;">91.2</td><td>46s</td><td style="text-align: right;"><b>91.3</b></td></tr>
<tr><td>amazon polarity</td><td style="text-align: center;">2</td><td style="text-align: right;">52s</td><td style="text-align: right;"><b>94.6</b></td><td>68s</td><td style="text-align: right;">94.2</td></tr>
<tr><td>dbpedia</td><td style="text-align: center;">1</td><td style="text-align: right;"><br /></td><td style="text-align: right;">98.1</td><td>8s</td><td style="text-align: right;"><b>98.4</b></td></tr>
<tr><td>dbpedia</td><td style="text-align: center;">2</td><td style="text-align: right;">8s</td><td style="text-align: right;">98.6</td><td>17s</td><td style="text-align: right;"><b>98.7</b></td></tr>
<tr><td>sogou news</td><td style="text-align: center;">1</td><td style="text-align: right;"><br /></td><td style="text-align: right;"><b>93.9</b></td><td>25s</td><td style="text-align: right;">93.6</td></tr>
<tr><td>sogou news</td><td style="text-align: center;">2</td><td style="text-align: right;">36s</td><td style="text-align: right;">96.8</td><td>30s</td><td style="text-align: right;"><b>96.9</b></td></tr>
<tr><td>yahoo answers</td><td style="text-align: center;">1</td><td style="text-align: right;"><br /></td><td style="text-align: right;"><b>72.0</b></td><td>30s</td><td style="text-align: right;">70.6</td></tr>
<tr><td>yahoo answers</td><td style="text-align: center;">2</td><td style="text-align: right;">27s</td><td style="text-align: right;"><b>72.3</b></td><td>48s</td><td style="text-align: right;">71.0</td></tr>
<tr><td>yelp full</td><td style="text-align: center;">1</td><td style="text-align: right;"><br /></td><td style="text-align: right;"><b>60.4</b></td><td>16s</td><td style="text-align: right;">56.9</td></tr>
<tr><td>yelp full</td><td style="text-align: center;">2</td><td style="text-align: right;">18s</td><td style="text-align: right;"><b>63.9</b></td><td>37s</td><td style="text-align: right;">60.0</td></tr>
<tr><td>yelp polarity</td><td style="text-align: center;">1</td><td style="text-align: right;"><br /></td><td style="text-align: right;"><b>93.8</b></td><td>10s</td><td style="text-align: right;">93.6</td></tr>
<tr><td>yelp polarity</td><td style="text-align: center;">2</td><td style="text-align: right;">15s</td><td style="text-align: right;"><b>95.7</b></td><td>20s</td><td style="text-align: right;">95.5</td></tr>
</tbody></table>
<br />
(Average accuracy for fastText is 83.2; for vw is 82.2.)<br />
<br />
In terms of accuracy, the two are roughly on par. vw occasionally wins; when it does, it's usually by 0.1% to 0.5%. fastText wins a bit more often, and on one dataset it wins significantly (yelp full: winning by 3%-4%) and on one a bit less (yahoo answers, up by about 1.3%). But the numbers are pretty much in line, and could almost certainly be brought up for vw with a wee bit of hyperparameter tuning (namely the learning rate, which is tuned in fastText).<br />
<br />
In terms of training time, fastText is maybe 30% faster on average, though these are such small datasets (eg 500k examples) that a difference of 52s versus 68s is not too significant. I also noticed that for most of the datasets, simply writing the model to disk for vw took a nontrivial amount of time. But wait, there's more. That 30% faster for fastText was <b>run on 20 cores in parallel</b> whereas the vw run did not use parallelized learning (vw runs two threads, one for I/O and one for learning).<br />
That said, a <b>major caveat</b> on comparing the training times. They're run on different machines. I don't know what type of machine the fastText results were achieved on, but it was a parallel 20-core run. The vw experiments were run on a single core, one pass over the data, on a 3.1Ghz Core i5-2400. Yes, I could have hogwild-ed vw and gotten it faster but it really didn't seem worth it for datasets this small. And yes, I could've rerun fastText on my machine, but... what can I say? I'm lazy.<br />
<br />
What did I do to get these vw numbers? Here's the entire training script:<br />
<blockquote class="tr_bq">
<pre>% cat run.sh
#!/bin/bash
d=$1
for ngram in 1 2 ; do
cat $d/train.csv | ./csv2vw.pl | \
time vowpal_wabbit/vowpalwabbit/vw --oaa `cat $d/classes.txt | wc -l` \
-b25 --ngram $ngram -f $d/model.$ngram
cat $d/test.csv | ./csv2vw.pl | \
time vowpal_wabbit/vowpalwabbit/vw -t -i $d/model.$ngram
done
</pre>
</blockquote>
Basically the only flags to vw are (1) telling it to do multiclass classification with one-against-all, (2) telling it to use 25 bits (not tuned), and telling it to either use unigrams or bigrams. [Comparison note: this means vw is using 33m hash bins; fastText used 10m for unigram models and 100m for bigram models.]<br />
<br />
The only(*) data munging that occurs is in csv2vw.pl, which is a lightweight script for converting the data, lowercasing, and doing very minor tokenization:<br />
<blockquote class="tr_bq">
<pre>% cat csv2vw.pl
#!/usr/bin/perl -w
use strict;
while (<>) {
chomp;
if (/^"*([0-9]+)"*,"(.+)"*$/) {
print $1 . ' | ';
$_ = lc($2);
s/","/ /g;
s/""/"/g;
s/([^a-z0-9 -\\]+)/ $1 /g;
s/:/C/g;
s/\|/P/g;
print $_ . "\n";
} else {
die "malformed line '$_'";
}
}
</pre>
</blockquote>
There are two exceptions where I did slightly more data munging. The datasets released for dbpedia and Soguo were not properly shuffled, which makes online learning hard. I preprocessed the training data by randomly shuffling it. This took 2.4s for dbpedia and 12s for Soguo.<br />
<br />
<b>[[[EDIT 2:20p 5 Aug 2016: </b>Out of curiosity, I upped the number of bits that vw uses for the experiments to 27 (so that it's on par with the 100m used by fastText). This makes it take about 5 seconds longer to run (writing the model to disk is slower). Performance stays the same on: ag news, amazon polarity, dbpedia, sogou, and yelp polarity; and it goes up from from 53.6/56.6 to 55.0/58.8 on amazon full, from 70.6/71.0 to 71.1/71.6 on yahoo answers, from 56.9/60.0 to 58.5/61.6 on yelp full. This puts the vw average with more bits at 82.6, which is 0.6% behind the fastText average.<b>]]]</b><br />
<br />
Long story short... am I switching from vw to fastText? Probably not any time soon.halhttp://www.blogger.com/profile/02162908373916390369noreply@blogger.com6tag:blogger.com,1999:blog-19803222.post-54535157527490617032016-07-29T11:18:00.002-06:002016-07-29T11:18:29.705-06:00A quick comment on structured input vs structured output learningWhen I think of structured input models, I typically think of things like kernels over discrete input spaces. For instance, the famous <a href="http://www.stat.purdue.edu/~vishy/talks/StringKernels.pdf">all-substrings kernel</a> for which K(d1,d2) effectively counts the number of common substrings in two documents, without spending exponential time enumerating them all. Of course there are many more ways of thinking about structured inputs: tree-to-string machine translation has a tree structured input. RNNs (on the input side) are essentially structured input models for sequence structures.<br />
<br />
When I think of structured output models, I typically think of things like CRFs, structured SVMs/M3Ns, multilabel predictors (those are borderline), various transition-based methods (eg., shift/reduce parsers), etc. Here, my internal model for the structure is essentially at prediction time: find a high scoring structure from this complicated discrete output space.<br />
<br />
Perhaps this has been obvious to everyone-but-me for a decade, but I only recently came to the recognition that these are essentially the same, at least if you restrict the sort of models you're willing to consider. (In particular, if you ignore things like imitation learning/learning to search for a moment.)<br />
<br />
In a pure structured input setting, you have some simple label space <i>Y</i> (let's assume it's the real numbers) and some complex input space<i> X</i>. Typically you want to learn a function <i>f : X ➝ Y</i>, which has low loss. In particular you want to minimize the expectation of <i>loss(y, f(x))</i> over random draws of <i>x,y</i>. And the "interesting" thing is that <i>x</i> isn't just a vector, so you have to be clever.<br />
<br />
In the pure structure output setting, in, for instance, the structured SVM/CRF setup, you have some input space <i>X</i> (which may or may not be structured) and some complex output space <i>Y</i>. As before, you want to learn a function <i>f : X ➝ Y</i>, which has low loss. However, in the most common setups, the way you accomplish this is that <i>instead</i> of directly learning <i>f</i>, you instead learn a <i>scoring</i> function <i>s</i> that scores <i>x,y</i> pairs based on how "good" that <i>y</i> is for the corresponding <i>x</i>. For a fixed scoring function <i>s</i>, you derive <i>f</i> according to the argmax rule: <i>f<sub>s</sub>(x) := argmax<sub>y</sub> s(x,y)</i>. In this way, you have effectively separated the learning problem (get a good <i>s</i>) from the structured problem (solve the argmax). [Whether this is good or not is up to debate; I'm personally on the "nay" side.] You then want to minimize something like the expectation of <i>loss(y, argmax<sub>y'</sub> s(x,y'))</i> over random draws <i>x,y</i>.<br />
<br />
The observation is that these two problems are essentially the same thing. That is, if you know how to do the structured input problem, then the structured output problem is essentially the same thing, as far as the learning problem goes. That is, if you can put structure in <i>f(x)</i> for structured input, you can just as well put structure in <i>s(x,y</i>)<i> </i>for structured output. Or, by example, if you can predict the fluency of an English sentence <i>x</i> as a structured input problem, you can predict the translation quality of a French/English sentence pair <i>x,y</i> in a structured output problem. This doesn't solve the <i>argmax</i> problem -- you have to do that separately -- but the underlying learning problem is essentially identical.<br />
<br />
You see similar ideas being reborn these days with papers like David Belanger's <a href="http://www.cs.umass.edu/~belanger/belanger_spen_icml.pdf">ICML paper this year on energy networks</a>. With this framework of think-of-structured-input-and-structured-output-as-the-same, basically what they're doing is building a structured score function that uses both the input and output simultaneously, and throwing these through a deep network. (Ok it's a bit more than that, but that's the cartoon.)<br />
<br />
At any rate, maybe obvious to everyone but me, but I thought I'd write it down anyway :).halhttp://www.blogger.com/profile/02162908373916390369noreply@blogger.com1tag:blogger.com,1999:blog-19803222.post-78648347709555344762016-07-26T14:42:00.002-06:002016-07-26T18:26:41.802-06:00Decoding (neural?) representationsI remember back in grad school days some subset of the field was thinking about the following question. I train an unsupervised HMM on some language data to get something-like-part-of-speech tags out. And naturally the question arises: these tags that come out... what are they actually encoding?<br />
<br />
At the time, there were essentially three ways of approaching this question that I knew about:<br />
<ol>
<li>Do a head-to-head comparison, in which you build an offline matching between induced tags and "true" tags, and then evaluate the accuracy of that matching. This was the standard evaluation strategy for unsupervised POS tagging, but is really just trying to get at the question of: how correlated are the induced tags with what we hope comes out.</li>
<li>Take a system that expects true POS tags and give it induced POS tags instead (at both training and test time). See how much it suffers (if at all). Joshua Goodman told me a few times (though I can't find his paper on this) that word clusters were just as good as POS tags if your task was NER.</li>
<li>Do something like #2, but also give the system <i>both</i> POS tags <i>and</i> induced tags, and see if the POS tags give you anything above and beyond the induced tags.</li>
</ol>
Now, ten years later since we're in "everything old is new again" phase, we're going through the same exercises, but with word embeddings instead of induced tags. This makes things slightly more complicated because it means that we have to have mechanisms that deal with continuous representations rather than discrete representations, but you basically see the same ideas floating around.<br />
<br />
In fact, of the above approaches, the only one that requires <i>any</i> modification is #1 because there's not an obvious way to do the matching. The alternative is to let a classifier do the matching, rather than an offline process. In particular, you take your embeddings, and then try to train a classifier that predicts POS tags from the embeddings directly. (Note: I claim this generalizes #1 because if you did this with discrete tags, the classifier would simply learn to do the matching that we used to compute "by hand" offline.) If your classifier can do a good job, then you're happy.<br />
<br />
This approach naturally has flaws (all do), but I think it's worth thinking about this seriously. To do so, we have to take a step back and ask ourselves: what are we trying to do? Typically, it seems we want to make an argument that a system that was not (obviously) designed to encode some phenomenon (like POS tags) and was not trained (specifically) to predict that phenomenon has nonetheless managed to infer that structure. (We then typically go on to say something like "who needs no POS tags" even though we just demonstrated our belief that they're meaningful by evaluating them... but okay.)<br />
<br />
As a first observation, there is an entire field of study dedicated to answering questions like this: (psycho)linguists. Admittedly they only answer questions like this in humans and not in machines, but if you've ever posed to yourself the question "do humans encode/represented phrase structures in their brains" and don't know the answer (or if you've never thought about this question!) then you should go talk to some linguists. More classical linguists would answer these questions with tests like, for instance, constituency tests or scoping tests. I like <a href="http://www.colinphillips.net/">Colin Phillips</a>' <a href="http://hal3.name/courses/2012F_CL1/out/phillips2003-syntax.pdf">encyclopedia article on syntax</a> for a gentle introduction (and is what I start with for syntax in intro NLP).<br />
<br />
So, as a starting point for "has my system learned X" we might ask our linguist friends how they determine if a human has learned X. Some techniques are difficult to replicate in machine (e.g., eye movement experiments, though of course models that have something akin to alignment---or "attention" if you must---could be thought of as having something like eye movements, though I would be hesitant to take this analogy too far). But many are not, for instance behavioral experiments, analyzing errors, and, I hesitate somewhat to say it, grammaticality judgements.<br />
<br />
My second comment has to do with the notion of "can these encodings be used to predict POS tags." Suppose the answer is "yes." What does that mean? Suppose the answer is "no."<br />
<br />
In order to interpret the answer to these questions, we have to get a bit more formal. We're going to train a classifier to do something like "predict POS given embedding." Okay, so what hypothesis space does that classifier have access to? Perhaps you say it gets a linear hypothesis space, in which case I ask: if it fails, why is that useful? It just means that POS cannot be decoded linearly from this encoding. Perhaps you make the hypothesis space outrageously complicated, in which case I ask: if it succeeds, what does that tell us?<br />
<br />
The reason I ask these questions is because I think it's useful to think about two extreme cases.<br />
<ol>
<li>We know that we can embed 200k words in about 300 dimensions with nearly orthogonal vectors. This means that for all intents and purposes, if we wanted, we could consider ourselves to be working with a one-hot word representation. We know that, to some degree, POS tags are predictable from words, especially if we allow for complex hypothesis spaces. But this is uninteresting because by any reasonable account, this representation has <i>not</i> encoded anything interesting: it's just the output classifier that's doing something interesting. That is to say: if your test can do well on the raw words as input, then it's dubious as a test.</li>
<li>We also know that some things are just unpredictable. Suppose I had a representation that perfectly encoded everything I could possibly want. But then in the "last layer" it got run through some encryption protocol. All of the information is still there, so the representation in some sense "contains" the POS tags, but no classifier is going to be able to extract it. That is to say, just because the encoded isn't on the "surface" doesn't mean it's not there. Now, one could reasonably argue something like "well if the information is there in an impossible-to-decode format then it might as well not be there" but this slope gets slippery very quickly.</li>
</ol>
Currently, I much prefer to think about essentially the equivalent of "behavioral" experiments. For instance, if you're machine translating and want to know if your system can handle scoping, then give it a minimal pair to translate that only differs in the scoping of some negation. Or if you're interesting in knowing whether it knows about POS tags, perhaps look at errors in its output and see if they fall along POS categories.<br />
<br />
<b>EDIT 26 Jul 2016, 8:24p Eastern:</b> It's unclear to a few people so clarification. I'm mostly <i>not</i> talking about type-level word embeddings above, but embeddings in context. At a type-level, you could imagine evaluating (1) on out of vocabular terms, which would be totally reasonable. I'm think more something like: the state of your biLSTM in a neural MT system. The issue is that if, for instance, this biLSTM can repredict the input (as in an autoencoder), then it could be that the POS tagger is doing all the work. See <a href="https://twitter.com/yoavgo/status/758051733595652096">this conversation thread</a> with Yoav Goldberg for some discussion.halhttp://www.blogger.com/profile/02162908373916390369noreply@blogger.com6tag:blogger.com,1999:blog-19803222.post-69018093334585881582016-07-12T15:21:00.003-06:002016-07-12T15:21:57.555-06:00Some picks from NAACL 2016Usual caveats: didn't see all talks, didn't read all papers, there's lot of good stuff at <a href="http://aclweb.org/anthology/N/N16/">NAACL</a> that isn't listed here! That said, here are some papers I particularly liked at NAACL, with some comments. Please add comments with papers you liked!<br />
<h4>
<a href="http://aclweb.org/anthology/N/N16/N16-1010.pdf">Automatic Summarization of Student Course Feedback</a> by Luo, Liu, Liu and Litman.</h4>
<br />
<br />
<br />
Anyone who has taught has suffered the following dilemma. You ask students for feedback throughout the course, and you have to provide free text because if you could anticipate their problems, you'd have addressed them already. But now you have hundreds of responses that you can't quickly read through. This paper provides a dataset of such responses, together with summaries that you actually have time to read through. The approach is a reasonably standard ILP formulation, with some additional machinery to deal with the fact that the data is very sparse. The idea is to essentially induce a "bigram similarity" as part of the summarization problem. I like this paper because the problem is great (I think NLP should really push in the direction of helping learners learn and teachers teach!), the dataset is nice (if a bit small) and the approach makes sense. And they actually do a human evaluation, even for a short paper! Hooray!<br />
<br />
<h4>
<a href="http://aclweb.org/anthology/N/N16/N16-1037.pdf">A Latent Variable Recurrent Neural Network for Discourse Relation Language Models</a> by Ji, Haffari and Eisenstein.</h4>
This paper combines the nice rich hypothesis classes you get with the "anonymous" latent variables in neural networks with the modeling power that one gets from the ability to marginalize "structured" latent variables from classic graphical models land. This is applied to document language modeling, in which the latent variables are discourse relations (in the PDTB sense). The model works well both for language modeling and for predicting implicit discourse relations and dialogue acts. (And, as one should expect from a paper with Eisenstein as an author, there are statistical significance tests!)<br />
<br />
<h4>
<a href="http://aclweb.org/anthology/N/N16/N16-1068.pdf">Structured Prediction with Output Embeddings for Semantic Image Annotation</a> by Quattoni, Ramisa, Madhyastha, Simo-Serra and Moreno-Noguer.</h4>
If you want to label images with complex structured outputs, like <i>play(dog,grass)</i>, you quickly run in to sparsity problems on the output. The proposal here is to decompose the outputs into embeddings (kind of like <a href="http://svivek.com/research/publications/SrikumarMa2014.pdf">Vivek's work</a> and <a href="http://www.umiacs.umd.edu/~jags/pdfs/jags12Reranking.pdf">Jags' work</a>) and learning a bilinear model of inputs/outputs in that space. In general, I think there's a lot to be done in interesting modeling of structured output spaces, and this paper gives a new set of techniques in this space.<br />
<br />
<h4>
<a href="http://aclweb.org/anthology/N/N16/N16-1073.pdf">Deconstructing Complex Search Tasks: A Bayesian Nonparametric Approach for Extracting Sub-tasks</a> by Mehrotra, Bhattacharya and Yilmaz.</h4>
<br />
The problem considered here is the following: if I want to go to ACL, I need to register, book a flight, book a hotel, find some attractions to go to while skipping sessions, look up all the good vegan restaurants in Berlin (hah!), etc. My overall task is going to ACL, but there are a number of highly related but different subtasks. The challenge is to infer these subtasks from search logs, so that you can provide better search support. The model is a Chinese Restaurant Process with word embeddings used to measuring similarities. And look, another short paper with a human evaluation!<br />
<br />
<h4>
<a href="http://aclweb.org/anthology/N/N16/N16-1098.pdf">A Corpus and Cloze Evaluation for Deeper Understanding of Commonsense Stories</a> by Mostafazadeh, Chambers, He, Parikh, Batra, Vanderwende, Kohli and Allen.</h4>
<br />
This paper introduces the task of story cloze: given a story prefix, predict which of two endings is the "correct" (natural) one. Eg: "Jim got his first credit card in college. He didn't have a job so he bought everything on his card. After he graduated he amounted a $10,000 debt. Jim realized that he was foolish to spend so much money." Then you have to decide on the right ending: "Jim decided to device a plan for repayment." (correct) versus "Jim decided to open another credit card." (incorrect). The data set was quite well constructed (lots of checks), and a large collection of baseline models are run for comparison. The data is available. I like this paper for the task and the very very careful data set collection. Since this was establishing baselines, I would really liked to have seen error bars on the results so we can have more reasonable future comparisons. I'm really tempted to try to annotate some of these with plot units, but that was really really painful the first time around; but I feel like that's a good explanatory theory for a lot of the examples shown in the paper.<br />
<br />
<h4>
<a href="http://aclweb.org/anthology/N/N16/N16-1114.pdf">Learning Global Features for Coreference Resolution</a> by Wiseman, Rush and Shieber. </h4>
The basic idea here is to take a "local" coreference model, that makes decisions about assigning mentions to entities, and augment it with a "global" model that models the entire cluster of mentions. In past work, cluster-level features have been hard to define (e.g., what are the right features to extract over sets of arbitrary size?). What's the solution? Essentially, embed the clusters. I like this paper because I wanted to try to do something like this and Wiseman did it better than I would have. I think there's also an interesting interpretation of this work in terms of things like neural Turing machines/memory networks, in which we now have technology for learning to update memory (where "memory" here is the "clusters"). My only gripe is that, like most papers in the new wave of coreference, the older ML approaches have been somewhat forgotten (except for Vincent Ng who is unforgettable); I'm thinking in particular about the work on ACE, like that of Luo, Ittycheriah, Jing, Kambhatla and Roukos on <a href="http://www.anthology.aclweb.org/P/P04/P04-1018.pdf">Bell-tree coreference</a>, which I don't think gets enough notice these days for introducing a lot of the ideas that make up modern coref.<br />
<br />
<h4>
<a href="http://aclweb.org/anthology/N/N16/N16-1147.pdf">Visual Storytelling</a> by Huang, Ferraro, Mostafazadeh, Misra, Agrawal, Devlin, Girshick, He, Kohli, Batra, Zitnick, Parikh, Vanderwende, Galley and Mitchell.</h4>
The main observation in this paper is that how you caption images in a sequence is different from how you caption them in isolation. For instance, if you have a temporal sequence of images from, say, a wedding, the captions for them should tell a story. This paper primarily introduces a dataset and baseline approaches for addressing the problem in this dataset. I like this paper, even if you remove the image stuff, because it emphasizes the fact that stories have a different structure than just sequences of sentences. The image stuff gives a grounding that's interesting beyond that. One problem pointed out in the paper is that automatic metrics aren't great here, which is problematic, but not particularly surprising.<br />
<br />
<br />halhttp://www.blogger.com/profile/02162908373916390369noreply@blogger.com0tag:blogger.com,1999:blog-19803222.post-64909741612583790482016-07-05T06:42:00.000-06:002016-07-05T06:42:38.014-06:00Rating the quality of reviews, after the factGroan groan groan reviewers are horrible people. Not you and me. <i>Those other reviewers over there!</i><br />
<br />
<b>tldr: </b>In general we actually don't think our reviews are that bad, though of course it's easy to remember the bad ones. Author perception of review quality is colored by, but not determined by, the overall accept/reject decision and/or the overall score that review gave to the paper.<i><br /></i><br />
<br />
NIPS did an experiment a bunch of years ago (can't dig it up any more, now it's an urban legend) where they asked reviewers at, I think, the time of author feedback, to rate the reviews. The anecdotal story was that there was almost perfect correlation between "this is a good review" and "this review gave my paper a high score." Of course this is not super surprising, even if you get rid of "emotions," because presumably I like my paper and so any review that doesn't like it is flawed.<br />
<br />
For NAACL 2013, we did a similar experiment, but we asked authors for their responses several months <i>after</i> the fact (actually, even after the conference had taken place), at which point hopefully emotions had cooled a bit and they could look back at their reviews with a sort of fond recollection. We presented the contact author of each paper with the original review text for each of their reviews, but did not show them the original scores. We asked them on a standard Likert scale how <i>helpful</i> this review was, and how <i>informative</i> this review was.<br />
<br />
Because this was long after the fact, response rate was of course not 100%, and it was also biased toward authors of papers that were accepted. We got responses from 128 authors on a total of 138 papers (some papers had same contact authors), covering a total of about 397 reviews (roughly one per paper, but some short papers only had two, and some papers had four).<br />
<br />
All the plots below are restricted to this set of 138 papers, not to the full set of about 500.<br />
<br />
First, let's get a sense of the data. Here are the overall results for this entire set of 138 papers:<br />
<div class="separator" style="clear: both; text-align: center;">
<a href="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEjnoC3hN8gpOa3nAHl4239VnP5MEURDhUuHYRJJBK1IYQAS9j2ub9XhDjmUaYRvvGJ-Jb3fdKM2XccplSXPjEEFrtzMtufbIPr_dGF8yCmldx8Hi7r5-aTTAkRpXdNq2bewhBgnOQ/s1600/reviewfeedback-summary.png" imageanchor="1" style="margin-left: 1em; margin-right: 1em;"><img border="0" height="400" src="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEjnoC3hN8gpOa3nAHl4239VnP5MEURDhUuHYRJJBK1IYQAS9j2ub9XhDjmUaYRvvGJ-Jb3fdKM2XccplSXPjEEFrtzMtufbIPr_dGF8yCmldx8Hi7r5-aTTAkRpXdNq2bewhBgnOQ/s400/reviewfeedback-summary.png" width="381" /></a></div>
<br />
(Note that the numbers add up to 397, not 138, because this is counting per-review not per-paper.) The first row shows the accept/reject ratio. Since NAACL 2013 had an acceptance rate between 25% and 30%, obviously survey results are biased toward accepted papers, but we still have a healthy response rate from rejected papers.<br />
<br />
Overall, the vast majority (~80%) of reviews were considered both informative and helpful (score 4 or 5) according to the authors. So yes, we need to do something about the 20% of reviews that got a 1, 2 or 3 on the Likert scale, but we're actually not doing that horribly. (Modulo sample selection bias.) The papers themselves were considered overwhelmingly appropriate and clear. The overall score distribution matches (roughly) the overall score distribution for the entire conference.<br />
<br />
Let's look at what happens if we look only at accepted or rejected paper:<br />
<div class="separator" style="clear: both; text-align: center;">
<a href="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEijEBSWwJw-UU66ABSLOgvbWeGGEfIvgPdKE2U9faSVhKpuY57S7WIzlwUfSKKVmg3KLk8egCJrixmWYZjTaurT8FwiXN6O4AnMfeP274H6SfTf0pKcHaMOjDUMcW3j3CPOJn_U8g/s1600/reviewfeedback-conditioned-Accepted+Papers.png" imageanchor="1" style="margin-left: 1em; margin-right: 1em;"><img border="0" height="400" src="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEijEBSWwJw-UU66ABSLOgvbWeGGEfIvgPdKE2U9faSVhKpuY57S7WIzlwUfSKKVmg3KLk8egCJrixmWYZjTaurT8FwiXN6O4AnMfeP274H6SfTf0pKcHaMOjDUMcW3j3CPOJn_U8g/s400/reviewfeedback-conditioned-Accepted+Papers.png" width="380" /></a></div>
<br />
<div class="separator" style="clear: both; text-align: center;">
<a href="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEjFn3NLVS8oMdnI9vNpeYeUT1TyQhEdBnyogoAVWfSszFptaisK3YdR4pBbuFzyuMRaLetagNQTLt7-LMUklpF4E0OrukldTyZy8gidgsjkERhKAqAf45H1Q2r1cR2diTdwaJCp4w/s1600/reviewfeedback-conditioned-Rejected+Papers.png" imageanchor="1" style="margin-left: 1em; margin-right: 1em;"><img border="0" height="400" src="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEjFn3NLVS8oMdnI9vNpeYeUT1TyQhEdBnyogoAVWfSszFptaisK3YdR4pBbuFzyuMRaLetagNQTLt7-LMUklpF4E0OrukldTyZy8gidgsjkERhKAqAf45H1Q2r1cR2diTdwaJCp4w/s400/reviewfeedback-conditioned-Rejected+Papers.png" width="380" /></a></div>
Comparing these, we definitely see a bit of the NIPS effect. For accepted papers, the reviews were considered overwhelmingly informative and helpful (scores of 4 or 5 in 85% or more cases). However, for rejected papers, the reviews were still considered largely informative and helpful (~73% of cases were 4s and 5s). Not surprisingly, accepted papers fare quite well on the individual score metrics, in particular overall score (duh!).<br />
<br />
We can alternatively condition the analysis on the overall paper score rather than the final accept/reject decision. Here's how that looks:<br />
<div class="separator" style="clear: both; text-align: center;">
<a href="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEgxjTkfExsZ15Th2eVsZcbZ6dHwHc7A_N_UpKxtfzXtnTYqIO45GvBMI4x9g8XqX-hpUG93FIF2btpM5F5qOYNX1JQ67WA4FDCimoHmHV9GSLsvceYiwyJq-TvWY9j95Lh1_w_Z5g/s1600/reviewfeedback-conditioned-High+Overall+Score.png" imageanchor="1" style="margin-left: 1em; margin-right: 1em;"><img border="0" height="400" src="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEgxjTkfExsZ15Th2eVsZcbZ6dHwHc7A_N_UpKxtfzXtnTYqIO45GvBMI4x9g8XqX-hpUG93FIF2btpM5F5qOYNX1JQ67WA4FDCimoHmHV9GSLsvceYiwyJq-TvWY9j95Lh1_w_Z5g/s400/reviewfeedback-conditioned-High+Overall+Score.png" width="381" /></a></div>
<br />
<div class="separator" style="clear: both; text-align: center;">
<a href="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEgzK3kswuKIMQDTXIa1CVc34TOCp5DhmcNzmoZCcp9RiiAYUe-wL16OiGP0wIgWMo7LfKr84_mAmSOxobFo18rQmNFyIbJl_D0VnXir6X_rl1FhlPnBKk90wT8frpVqPcW5QZxCOQ/s1600/reviewfeedback-conditioned-Low+Overall+Score.png" imageanchor="1" style="margin-left: 1em; margin-right: 1em;"><img border="0" height="400" src="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEgzK3kswuKIMQDTXIa1CVc34TOCp5DhmcNzmoZCcp9RiiAYUe-wL16OiGP0wIgWMo7LfKr84_mAmSOxobFo18rQmNFyIbJl_D0VnXir6X_rl1FhlPnBKk90wT8frpVqPcW5QZxCOQ/s400/reviewfeedback-conditioned-Low+Overall+Score.png" width="381" /></a></div>
That's not substantially different.<br />
<br />
So what makes the difference between good (informativeness 4 or 5) and bad (informativeness 1 or 2) reviews?<br />
<br />
On average, the "good" reviews were about 15% longer than the bad reviews (on average 320 characters versus 280 characters).<br />
<br />
Somewhat surprisingly, a linear classifier on bag of words data and distinguish with 90% accuracy "good" from "bad" reviews, but the features it gives high weight to are basically features that look like positive versus negative reviews, rather essentially exploiting of the correlation between informativeness and acceptance, rather than informativeness on its own.halhttp://www.blogger.com/profile/02162908373916390369noreply@blogger.com3