On the Relationship Between Adversarial Robustness and Decision Region in Deep Neural Networks

Deep Neural Networks (DNNs) are increasingly recognized for their susceptibility to various adversarial attacks. To enhance their robustness, adversarial training (AT) has been a popular approach. However, traditional AT methods often focus on a limited range of attack types, leaving DNNs vulnerable to unforeseen threats. This paper introduces Calibrated Adversarial Sampling (CAS), a novel fine-tuning technique that employs a multi-armed bandit framework to optimize DNNs against a broader spectrum of adversarial perturbations, thereby improving their overall resilience.

The paper titled 'FQ-PETR: Fully Quantized Position Embedding Transformation for Multi-View 3D Object Detection' addresses the challenges of deploying PETR models in autonomous driving due to their high computational costs and memory requirements. It introduces FQ-PETR, a fully quantized framework that aims to enhance efficiency without sacrificing accuracy. Key innovations include a Quantization-Friendly LiDAR-ray Position Embedding and techniques to mitigate accuracy degradation typically associated with quantization methods.