Multi-view Representative and Informative Induced Active Learning

Abstract

Most existing active learning methods often manually label samples and train models with labeled data in an iterative way. Unfortunately, at the early stage of the experiment, few labeled data are available, hence, selecting the most valuable data points to label is necessary and important. To this end, we propose a novel method, called Multi-view Representative and Informative-induced Active Learning ( MRI-AL ), which selects samples of both representativeness and informativeness with the help of complementarity of multiple views. Specifically, subspace reconstruction with structure sparsity technique is employed to ensure the selected samples to be representative, while the global similarity constraint guarantees the informativeness of the selected samples. The proposed method is solved efficiently by alternating direction method of multipliers (ADMM). We empirically show that our method outperforms existing early experimental design approaches.