Timezone: »
Deep networks are well-known to be fragile to adversarial attacks. We conduct an empirical analysis of deep representations under the state-of-the-art attack method called PGD, and find that the attack causes the internal representation to shift closer to the ``false'' class. Motivated by this observation, we propose to regularize the representation space under attack with metric learning to produce more robust classifiers. By carefully sampling examples for metric learning, our learned representation not only increases robustness, but also detects previously unseen adversarial samples. Quantitative experiments show improvement of robustness accuracy by up to 4% and detection efficiency by up to 6% according to Area Under Curve score over prior work. The code of our work is available at https://github.com/columbia/MetricLearningAdversarial_Robustness.
Author Information
Chengzhi Mao (Columbia University)
Ziyuan Zhong (Columbia University)
Junfeng Yang (Columbia University)
Carl Vondrick (Columbia University)
Baishakhi Ray (Columbia University)
More from the Same Authors
-
2022 Poster: Representing Spatial Trajectories as Distributions »
Didac Suris Coll-Vinent · Carl Vondrick -
2022 Poster: Private Multiparty Perception for Navigation »
Hui Lu · Mia Chiquier · Carl Vondrick -
2020 Poster: Listening to Sounds of Silence for Speech Denoising »
Ruilin Xu · Rundi Wu · Yuko Ishiwaka · Carl Vondrick · Changxi Zheng -
2019 Poster: Noise-tolerant fair classification »
Alex Lamy · Ziyuan Zhong · Aditya Menon · Nakul Verma -
2018 Poster: Efficient Formal Safety Analysis of Neural Networks »
Shiqi Wang · Kexin Pei · Justin Whitehouse · Junfeng Yang · Suman Jana