Learning with a Strong Adversary

CoRR, Volume abs/1511.03034, 2015.

Cited by: 186|Views120
EI
Weibo:
The experimental results suggests that learning with a strong adversary is promising in the sense that compared to the benchmarks in the literature, it achieves significantly better robustness while maintain high normal accuracy

Abstract:

The robustness of neural networks to intended perturbations has recently attracted significant attention. In this paper, we propose a new method, \emph{learning with a strong adversary}, that learns robust classifiers from supervised data. The proposed method takes finding adversarial examples as an intermediate step. A new and simple w...More

Code:

Data:

0
Your rating :
0

 

Tags
Comments