Leveraging Ensemble Diversity for Robust Self-Training in the presence of Sample Selection Bias
Ambroise Odonnat
Title: Leveraging Ensemble Diversity for Robust Self-Training in the presence of Sample Selection Bias
Abstract: Self-training is a well-known approach for semi-supervised learning. It consists of iteratively assigning pseudo-labels to unlabeled data for which the model is confident and treating them as labeled examples. For neural networks, softmax prediction probabilities are often used as a confidence measure, although they are known to be overconfident, even for wrong predictions. This phenomenon is particularly intensified in the presence of sample selection bias, i.e., when data labeling is subject to some constraint. In this talk, we introduce a novel confidence measure built upon the prediction diversity of an ensemble of linear classifiers. We provide a theoretical analysis to characterise the relationship between the diversity of the ensemble’s classifiers and their performance. We empirically demonstrate the calibration of our confidence measure and its benefits for three different pseudo-labeling policies on classification datasets of various data modalities.
Website: https://scholar.google.com/citations?hl=en&user=M_OS-3kAAAAJ
In Room M7.101 of Monod campus, ENSL.