Active Learning by Increasing Model Likelihood for Gaussian Mixture Models Classifiers Bambang Heru Iswanto
Universitas Negeri Jakarta
Abstract
A crucial problem in many classification tasks is to acquire labeled data, while a large amounts of unlabeled data are available. One way to overcome these problems is to apply an active learning process. This technique aims to select the most valuable examples or instances for labeling and to build an optimal generative classifier. In this paper, we propose an active learning algorithm for Gaussian mixture model classifiers to select the most informative examples called the increasing model likelihood (IML). The method assumes a large pool of unlabeled examples be available and selects an instance for labeling that maximize likelihood of the model. Simulation and experimental results show that the proposed method outperforms the passive learning method, can reduce the labeling cost over 60% on binary datasets, and more robust in handling noisy data.
Keywords: Active Learning, Likelihood, Gaussian Mixture Models, Classification
