Using Early Readouts to Mediate Featural Bias in Distillation
Deep networks tend to learn spurious feature-label correlations in real-world supervised learning tasks. This vulnerability is aggravated in distillation, where a (student) model may have less representational capacity than the corresponding teacher model. Often, knowledge of specific problem features is used to reweight instances & rebalance the learning process. We propose a novel early readout mechanism whereby we attempt to predict the label using representations from earlier network layers. We show that these early readouts automatically identify problem instances or groups in the form of confident, incorrect predictions. We improve group fairness measures across benchmark datasets by leveraging these signals to mediate between teacher logits and supervised label. We extend our results to the closely related but distinct problem of domain generalization, which also critically depends on the quality of learned features. We provide secondary analyses that bring insight into the role of feature learning in supervision and distillation.