Vanilla Feature Distillation for Improving the Accuracy-Robustness Trade-Off in Adversarial Training.
Guodong CaoZhibo WangXiaowei DongZhifei ZhangHengchang GuoZhan QinKui RenPublished in: CoRR (2022)
Keyphrases
- trade off
- high accuracy
- computational efficiency
- bias variance
- prediction accuracy
- supervised learning
- test set
- classification accuracy
- error rate
- training and testing data
- highly accurate
- computational complexity
- training samples
- computational cost
- feature vectors
- artificial neural networks
- training phase
- discrimination power
- numerical stability
- classification performances
- face recognition
- database