tag:blogger.com,1999:blog-19803222.post726284263469295772..comments2024-03-18T01:45:45.724-06:00Comments on natural language processing blog: Multi-task learning: should our hypothesis classes be the same?halhttp://www.blogger.com/profile/02162908373916390369noreply@blogger.comBlogger5125tag:blogger.com,1999:blog-19803222.post-39277589590709921892010-09-03T07:13:08.877-06:002010-09-03T07:13:08.877-06:00Maybe something like an HDP is right here? Not in ...Maybe something like an HDP is right here? Not in the usual setting, but something that assumes the feature,weight pairs are drawn from a DP for each domain with an HDP prior, and then you can sample/optimize which are the specific features for each domain and which share weights.<br /><br />Inference would be tricky, but I think the idea may be sound.Alexandre Passoshttps://www.blogger.com/profile/10099321916600547808noreply@blogger.comtag:blogger.com,1999:blog-19803222.post-10019357227430639452010-08-25T22:52:04.590-06:002010-08-25T22:52:04.590-06:00Yeah, it sounds very reasonable to want to have di...Yeah, it sounds very reasonable to want to have different feature spaces for different tasks. Is there any theory in ML that depends on them being in the same hypothesis class? <br /><br />I think there is some work on adaptation with different feature spaces (via feature extraction). For example: <a href="http://keg.cs.tsinghua.edu.cn/persons/tj/publications/CIKM09-Wang-Tang-et-al-HCDRank.pdf" rel="nofollow">Heterogenous cross-domain ranking [Wang09]</a> frames the learning problem as (simplified here): <br /><br />min_{Ws,Wt,U} L(Ws,U*Xs) + L(Wt,U*Xt) where Xs and Xt are source and target samples, Ws and Wt are the source and target weights to be learned, L is the loss function, and U is a transformation mapping Xs and Xt (which could be different spaces, into the same latent subspace. <br /><br />For practical reasons, they still need to concat that source and target feature vectors, or else matrix dimensions of U won't match. The optimization is done using a convex trick similar to <a href="http://books.nips.cc/papers/files/nips19/NIPS2006_0251.pdf" rel="nofollow">[Argyriou06]</a>. <br /><br />I think your CCA idea is very cool and possibly more effective, though!Kevin Duhhttps://www.blogger.com/profile/07407894290644783502noreply@blogger.comtag:blogger.com,1999:blog-19803222.post-33287061739808437802010-08-23T19:58:44.702-06:002010-08-23T19:58:44.702-06:00thanks
my work is to use domain adaptation into m...thanks <br />my work is to use domain adaptation into multitask learning and it is need to be incremental learning,so i am very appreciate if you have some idea about that.<br />thank you againcissahttp://cisseimpact@hotmail.co.jpnoreply@blogger.comtag:blogger.com,1999:blog-19803222.post-39376936230557843112010-08-22T17:19:46.020-06:002010-08-22T17:19:46.020-06:00@cissa: i hadn't seen that, but it's very ...@cissa: i hadn't seen that, but it's very related to a previous post: http://nlpers.blogspot.com/2008/05/adaptation-versus-adaptability.html<br /><br />btw, you can read the paper here: http://www.lib.kobe-u.ac.jp/repository/90001004.pdfhalhttps://www.blogger.com/profile/02162908373916390369noreply@blogger.comtag:blogger.com,1999:blog-19803222.post-89824126686935629802010-08-22T08:48:48.306-06:002010-08-22T08:48:48.306-06:00http://portal.acm.org/citation.cfm?id=1657504.1657...http://portal.acm.org/citation.cfm?id=1657504.1657510<br />can you read this paper,do you have some idea about incremental multitask learning?cissahttp://cisseimpact@hotmail.co.jpnoreply@blogger.com