Rethinking Evaluation Paradigms in IBP-based Certified Training 事件

Rethinking Evaluation Paradigms in IBP-based Certified Training arXiv:2606.02134v1 Announce Type: cross Abstract: Deep neural networks achieve strong performance on many supervised learning tasks but remain vulnerable to adversarial perturbations. Neural network verification provides mathematically rigorous robustness guarantees, yet at substantial computational cost. To mitigate this, certified training techniques optimise for verifiable robustness during training, typically inducing a trade-o