@inproceedings{DBLP:conf/icml/ZhangZ19,
author = {Zhang, Tianyuan and Zhu, Zhanxing},
booktitle = {Proceedings of the 36th International Conference on Machine Learning, {\{}ICML{\}} 2019, 9-15 June 2019, Long Beach, California, {\{}USA{\}}},
editor = {Chaudhuri, Kamalika and Salakhutdinov, Ruslan},
pages = {7502--7511},
publisher = {PMLR},
series = {Proceedings of Machine Learning Research},
title = {{Interpreting Adversarially Trained Convolutional Neural Networks}},
url = {http://proceedings.mlr.press/v97/zhang19s.html},
volume = {97},
year = {2019}
}
We find that AT-CNNs are better at captur-ing long-range correlations such as shapes, and less biased towards textures than normally trained CNNs in popular object recognition datasets.