关键词: 多任务分类, 罗杰斯特回归, 后验概率, 对偶坐标下降法

Abstract: When facing multi-task learning problems, it is desirable that the learning method can find the correct input-output features and share the commonality among multiple domains and also scale up for large multi-task datasets. This paper introduces the multi-task coupled logistic regression framework called MTC-LR, which is a new method for generating each classifier for each task, capable of sharing the commonality among multi-task domains. The basic idea of MTC-LR is to use all individual logistic regression based classifiers, each one appropriate for each task domain, but in contrast to other SVM based proposals, learning all the parameter vectors of all individual classifiers by using the conjugate gradient method, in a global way and without the use of kernel trick, and being easily extended into its scaled version. This paper theoretically shows that the addition of a new term in the cost function of the set of logistic regressions（that penalizes the diversity among multiple tasks） produces a coupling of multiple tasks that allows MTC-LR to improve the learning performance in a logistic-regression way. This finding can make us easily integrate it with a state-of-the-art fast logistic regression algorithm called CDdual to develop its fast version MTC-LR-CDdual for large multi-task datasets. The proposed algorithm MTC-LR-CDdual is also theoretically analyzed. The experimental results on artificial and real datasets indicate the effectiveness of the proposed algorithm MTC-LR-CDdual in classification accuracy, speed and robustness.

Key words: multi-task classification learning, logistic regression, posterior probability, dual coordinate descent method