Active learning with c-certainty
PAKDD'12 Proceedings of the 16th Pacific-Asia conference on Advances in Knowledge Discovery and Data Mining - Volume Part I
| Hi-index | 0.03 |
When active learning is applied to real-world applications, human experts usually act as oracles to provide labels. However, human make mistakes, thus noise might be introduced during the learning process. Most previous studies simplify the problem by assuming uniformly-distributed noise over the sample space. Such assumption, however, might fail to precisely reflect the human experts' behaviour in real-world situations. In this paper, we therefore study active learning with such human-like oracles, by making a more realistic assumption that the noise is example-dependent (i.e., non-uniformly distributed over the sample space). More specifically, when the human-like oracle is highly confident in labelling examples, it is naturally less likely to provide incorrect answers, whereas when such confidence is low, the noise would be more likely to be introduced. Based on the analysis of such human-like oracle, we propose a generic yet simple active learning algorithm to simultaneously explore the unlabelled data and exploit the labelled data. Empirical study on both synthetic and real-world data sets verifies the superiority of the proposed algorithm, compared with the traditional uncertainty sampling.