30 May 2009

Semi-supervised or Semi-unsupervised? (SSL-NLP invited papers)

Kevin Duh not-so-recently asked me to write a "position piece" for the workshop he's co-organizing on semi-supervised learning in NLP. I not-so-recently agreed. And recently I actually wrote said position piece. You can also find a link off the workshop page. I hope people recognize that it's intentionally a bit tongue-in-cheek. If you want to discuss this stuff or related things in general, come to the panel at NAACL from 4:25 to 5:25 on 4 June at the workshop! You can read the paper for more information, but my basic point is that we can typically divide semi-supervised approached into one lump (semi-supervised) that work reasonably well with only labeled data and are just improved with unlabeled data and one lump (semi-unsupervised) that work reasonably well with only unlabeled data and are just improved with labeled data. The former are typically encode lots of prior information; the latter do not. Let's combine! (Okay, my claim is more nuanced than that, but that's the high-order bit.)
at

11 comments:

  1. Rachel Cotterill01 June, 2009 10:36

    I was tempted to come to that workshop but I've signed up for the two semantics ones instead. Looks like I have a full week of semantics ahead, actually.

    Rachel ~ writing from the Ballroom in UMC!

    ReplyDelete
  2. Bob Carpenter01 June, 2009 13:16

    I don't see the generative/discriminative distinction as such a dichotomy. One attractive feature of Bayesian inference is that once you've set up a model, arbitrary subsets of the random variables may be supervised, which amounts to simply fixing their outcomes.

    With hierarchical Bayesian GLMs, one typically models (a) the hierarchical mean and variance parameters, (b) the regression coefficients, and (c) the predicted outcomes including noise. One cannot by definition model (d) the highest level priors. Typically, one does not model (e) the item level predictors.

    But really this choice is arbitrary. You can choose not to model the hierarchical variance parameters or you can set the hiearchical prior means to zero, for instance. For missing data imputation, you see the predictors themselves being modeled to fill in missing data (often with an ad-hoc regression based on the other predictors).

    ReplyDelete
  3. Anonymous24 October, 2009 01:40

    Very nice information. Thanks for this. Please come visit my site Fortworth Yellow Page Business Directory when you got time.

    ReplyDelete
  4. Anonymous24 October, 2009 01:40

    Very nice information. Thanks for this. Please come visit my site Local Business Directory Of Fortworth U.S.A. when you got time.

    ReplyDelete
  5. Anonymous30 October, 2009 22:33

    I enjoyed reading your work! GREAT post! I looked around for this… but I found you! Anyway, would you mind if I threw up a backlink from my site? Please come visit my site Phone Directory Of Columbus City Ohio OH State when you got time.

    ReplyDelete
  6. Anonymous30 October, 2009 22:34

    I enjoyed reading your work! GREAT post! I looked around for this… but I found you! Anyway, would you mind if I threw up a backlink from my site? Please come visit my site Business Directory Columbus when you got time.

    ReplyDelete
  7. Anonymous31 October, 2009 03:01

    the predicted outcomes including noise. One cannot by definition model (d) the highest level priors. Typically, one does not model (e) the item level predictors.
    Assignments | Theses | Essays

    ReplyDelete
  8. Anonymous31 October, 2009 03:01

    arbitrary subsets of the random variables may be supervised, which amounts to simply fixing their outcomes.
    A Level Coursework | Dissertations

    ReplyDelete
  9. Business Plan Presentation11 December, 2009 00:23

    Hey,
    Very useful piece of information indeed! Thanks for sharing and keep posting the good work.

    Business Plan Presentation

    ReplyDelete
  10. Anonymous14 April, 2010 01:34

    Latest film "Nightmare on Elm Street" Plot
    "The new Nightmare on Elm Street" played by the screenwriter Wesley Stricker, will follow the original director Wes Craven's Queen's routine, to stimulate a bloody killer scenes big screen again. According to reports, film the story will be a nightmare killer Freddy is how to burn the first incarnation of devils as the main history.

    We look forward to this wonderful movie please right..............................................
    New Nightmare on Elm Street "will be April 30, 2010 capture of the North American cinemas.

    ReplyDelete
  11. gamefan1215 April, 2010 10:03

    i would definitely go with the supervised. I think you have to go with this in the future.
    palm beach cosmetic dentist

    ReplyDelete