Robustifying Point Cloud Networks by Refocusing

arXiv:2308.05525v3 Announce Type: replace-cross
Abstract: The ability to cope with out-of-distribution (OOD) corruptions and adversarial attacks is crucial in real-world safety-demanding applications. In this study, we develop a general mechanism to increase neural network robustness based on focus analysis.
Recent studies have revealed the phenomenon of \textit{Overfocusing}, which leads to a performance drop. When the network is primarily influenced by small input regions, it becomes less robust and prone to misclassify under noise and corruptions.
However, quantifying overfocusing is still …

abstract adversarial adversarial attacks analysis applications arxiv attacks cloud cs.cv cs.lg distribution focus general leads network networks neural network performance robustness safety studies study type world