论文信息 - Cross Domain Active Learning

Cross Domain Active Learning

In this paper, we propose a solution to reduce the labeling costs by applying domain adaption methods coupled with active learning to reduce the number labels needed to train a classifier. We assume to have only one task but different domains in the sense that we have texts that come from different distributions. Our approach uses multi domain learning together with active learning to find a minimum number of texts to label from as few domains as possible to train a classifier with a certain confidence in its predictions.

Christian Pölitz | Christian Pölitz

[1] John Platt,et al. Probabilistic Outputs for Support vector Machines and Comparisons to Regularized Likelihood Methods , 1999 .

[2] A. Owen,et al. Safe and Effective Importance Sampling , 2000 .

[3] Burr Settles,et al. Active Learning Literature Survey , 2009 .

[4] Yi Lin,et al. Support Vector Machines for Classification in Nonstandard Situations , 2002, Machine Learning.

[5] Carolyn Penstein Rosé,et al. Multi-Domain Learning: When Do Domains Matter? , 2012, EMNLP-CoNLL.

[6] ChengXiang Zhai,et al. Instance Weighting for Domain Adaptation in NLP , 2007, ACL.

[7] Daniel Marcu,et al. Domain Adaptation for Statistical Classifiers , 2006, J. Artif. Intell. Res..

[8] David D. Lewis,et al. Heterogeneous Uncertainty Sampling for Supervised Learning , 1994, ICML.

[9] John Blitzer,et al. Domain Adaptation with Structural Correspondence Learning , 2006, EMNLP.

[10] Thorsten Joachims,et al. Learning to classify text using support vector machines - methods, theory and algorithms , 2002, The Kluwer international series in engineering and computer science.