Title: Active learning classification with variable selection
Authors: Yuan-chin Chang - Academia Sinica (Taiwan) [presenting]
Abstract: Active learning usually refers to certain kinds of learning methods that could select learning subjects sequentially. We will discuss some logistic model-based active learning methods with variable selection features in addition to subject selection strategies. We adopt a batch subject selection strategy with a modified sequential experimental design method and simultaneously conduct a greedy variable selection procedure such that we can update the classification model with all labeled training subjects. Another method is based on the stochastic regression where subjects are selected one-by-one adaptively and variables are identified sequentially. We repeat these algorithms repeatedly until a corresponding stopping criterion is reached. Our numerical results confirm that the proposed procedures can produce competitive performances with a smaller training size and a more compact model compared with that of the classifier trained with all variables and a full data set.