Proj CDeepFuzz Paper Reading: Invariance-inducing regularization using worst-case transformations suffices to boost accuracy and spatial robustness

Abstract

本文：
Task: 1. prove invariance-inducing regularizers can increase predictive accuracy for worst-case spatial transformations 2. prove that on adversarial examples from transformation groups in the infinite data limit robust training can also improve accuracy on test set
实验：
数据集：CIFAR-10, SVHN, CIFAR-100

  1. 证明在标准data augmentation或者adversarial training上添加regularization能够减少relative robust error 20%(CIFAR-10)且overhead最小
  2. 效果甚至好于人工调整过的模型
  3. 在SVHN上同时提高了test set的精确度
posted @ 2023-09-05 23:14  雪溯  阅读(4)  评论(0)  编辑  收藏  举报