Faster ISNet for Background Bias Mitigation on Deep Neural Networks

arXiv:2401.08409v2 Announce Type: replace-cross
Abstract: Bias or spurious correlations in image backgrounds can impact neural networks, causing shortcut learning (Clever Hans Effect) and hampering generalization to real-world data. ISNet, a recently introduced architecture, proposed the optimization of Layer-Wise Relevance Propagation (LRP, an explanation technique) heatmaps, to mitigate the influence of backgrounds on deep classifiers. However, ISNet's training time scales linearly with the number of classes in an application. Here, we propose reformulated architectures whose training time becomes independent from this …

abstract architecture arxiv bias correlations cs.cv cs.cy cs.lg data eess.iv faster hans image impact influence layer networks neural networks optimization propagation shortcut type wise world