NeurIPS 2019
Sun Dec 8th through Sat the 14th, 2019 at Vancouver Convention Center
Paper ID:1853
Title:Adversarial training for free!

The paper proposes using the backward pass of backprop algorithm to compute the adversarial example. This leads to computational time saving because instead of two backward passes. Overall, a good paper. The rebuttal was generally satisfying. It's encouraged to further consider the comments in the final version.