DeNetDM: Debiasing by Network Depth Modulation

arXiv:2403.19863v1 Announce Type: new
Abstract: When neural networks are trained on biased datasets, they tend to inadvertently learn spurious correlations, leading to challenges in achieving strong generalization and robustness. Current approaches to address such biases typically involve utilizing bias annotations, reweighting based on pseudo-bias labels, or enhancing diversity within bias-conflicting data points through augmentation techniques. We introduce DeNetDM, a novel debiasing method based on the observation that shallow neural networks prioritize learning core attributes, while deeper ones emphasize biases when …

abstract annotations arxiv augmentation bias biases challenges correlations cs.cv cs.lg current data datasets diversity labels learn network networks neural networks robustness through type