Adversarial Weight Perturbation Helps Robust Generalization(AWP adversarial train )
目录主要创新点实验探究Weightlosslandscape与robustgeneralizationgap之间的关系Weightlosslandscape的绘制在learningprocessingofvanillaAT中(即探究一个模型)不同对抗训练方式最终的weightlosslandscape对比(即探究多个模型)正式提出AdversarialWeightPerturbationAT-AW