Active Learning for Multi-class Logistic Regression Export

@inproceedings{citeulike:1650580, abstract = {Which of the many proposed methods for ac- tive learning can we expect to yield good per- formance in learning logistic regression clas- si¯ers? In this article, we evaluate di®erent approaches to determine suitable practices. Among our contributions, we test several ex- plicit objective functions for active learning: an empirical consideration lacking in the lit- erature until this point. We develop a theo- retical framework for applying di®erent loss functions motivated by work in optimal ex- perimental design. Empirical investigations demonstrate the bene¯ts of our variance re- duction method which gives attractive classi- ¯cation accuracy and matches or beats ran- dom performance in all evaluations. Of the alternative heuristic approaches, we identify a method called margin sampling as giving promising performance with little computa- tional overhead.}, author = {Andrew, I. and Schein and Ungar, Lyle H.}, citeulike-article-id = {1650580}, citeulike-linkout-0 = {http://citeseer.ist.psu.edu/rd/8088578%2C732620%2C1%2C0.25%2CDownload/http://citeseer.ist.psu.edu/cache/papers/cs2/36/http:zSzzSzwww.cis.upenn.eduzSz%7EaiszSz.zSzpublicationszSzicml2005.pdf/active-learning-for-multi.pdf}, keywords = {active-learning, logistic-regression, maximum-entropy}, posted-at = {2007-09-13 07:01:14}, priority = {3}, title = {Active Learning for Multi-class Logistic Regression}, url = {http://citeseer.ist.psu.edu/rd/8088578%2C732620%2C1%2C0.25%2CDownload/http://citeseer.ist.psu.edu/cache/papers/cs2/36/http:zSzzSzwww.cis.upenn.eduzSz%7EaiszSz.zSzpublicationszSzicml2005.pdf/active-learning-for-multi.pdf} }

TY - CONF ID - citeulike:1650580 L3 - citeulike-article-id:1650580 N2 - Which of the many proposed methods for ac- tive learning can we expect to yield good per- formance in learning logistic regression clas- si¯ers? In this article, we evaluate di®erent approaches to determine suitable practices. Among our contributions, we test several ex- plicit objective functions for active learning: an empirical consideration lacking in the lit- erature until this point. We develop a theo- retical framework for applying di®erent loss functions motivated by work in optimal ex- perimental design. Empirical investigations demonstrate the bene¯ts of our variance re- duction method which gives attractive classi- ¯cation accuracy and matches or beats ran- dom performance in all evaluations. Of the alternative heuristic approaches, we identify a method called margin sampling as giving promising performance with little computa- tional overhead. TI - Active Learning for Multi-class Logistic Regression KW - active-learning KW - logistic-regression KW - maximum-entropy AU - Andrew, I AU - Schein AU - Ungar, Lyle UR - http://citeseer.ist.psu.edu/rd/8088578%2C732620%2C1%2C0.25%2CDownload/http://citeseer.ist.psu.edu/cache/papers/cs2/36/http:zSzzSzwww.cis.upenn.eduzSz%7EaiszSz.zSzpublicationszSzicml2005.pdf/active-learning-for-multi.pdf ER -