Structured Gradient-based Interpretations via Norm-Regularized Adversarial Training

arXiv:2404.04647v1 Announce Type: new
Abstract: Gradient-based saliency maps have been widely used to explain the decisions of deep neural network classifiers. However, standard gradient-based interpretation maps, including the simple gradient and integrated gradient algorithms, often lack desired structures such as sparsity and connectedness in their application to real-world computer vision models. A frequently used approach to inducing sparsity structures into gradient-based saliency maps is to alter the simple gradient scheme using sparsification or norm-based regularization. A drawback with such post-processing …

abstract adversarial adversarial training algorithms application arxiv classifiers computer computer vision cs.cv decisions deep neural network gradient however interpretation maps network neural network norm simple sparsity standard training type via vision vision models world