Active Learning for Support Vector Machines with Maximum Model Change

Margin-based strategies and model change based strategies represent two important types of strategies for active learning. While margin-based strategies have been dominant for Support Vector Machines (SVMs), most methods are based on heuristics and lack a solid theoretical support. In this paper, we propose an active learning strategy for SVMs based on Maximum Model Change (MMC). The model change is defined as the difference between the current model parameters and the updated parameters obtained with the enlarged training set. Inspired by Stochastic Gradient Descent (SGD) update rule, we measure the change as the gradient of the loss at a candidate point. We analyze the convergence property of the proposed method, and show that the upper bound of label requests made by MMC is smaller than passive learning. Moreover, we connect the proposed MMC algorithm with the widely used simple margin method in order to provide a theoretical justification for margin-based strategies. Extensive experimental results on various benchmark data sets from UCI machine learning repository have demonstrated the effectiveness and efficiency of the proposed method.