A Closer Look at Accuracy vs. Robustness

By Yao-Yuan Yang, Cyrus Rashtchian, Hongyang Zhang, Ruslan Salakhutdinov, Kamalika Chaudhuri. Neural Information Processing Systems, 2020

328 citations recorded by the source corpus.

View Paper: https://www.semanticscholar.org/paper/ce7ae617528c1cdc2b1efb0010e8a8991e7c90b7

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

Citations in the corpus

Adversarial Robustness with Semi-Infinite Constrained Learning (2021, background): “Furthermore, adversarial training often degrades the performance of the model on clean data [52, 75, 76].”
RobustBench: a standardized adversarial robustness benchmark (2020, background): “Moreover, since prior 338 works [51, 146] connected higher smoothness with better robustness, we analyze the smoothness 339 of the models both at intermediate and output layers.”
Geometry-aware Instance-reweighted Adversarial Training (2020, methodology): “On the other hand, some argued that there is no such the trade-off: Raghunathan et al. (2020) showed infinite data could eliminate this trade-off; Yang et al. (2020) showed benchmark image datasets are class-separated. ar X iv :2 01 0.”
Unleashing the Power of Contrastive Self-Supervised Visual Models via Contrast-Regularized Fine-Tuning (2021, methodology): “To our knowledge, this is quite promising since even the most advanced AT methods [65, 69] ﬁnd it difﬁcult to overcome the accuracy-robustness trade-off [67].”
Recent Advances in Adversarial Training for Adversarial Robustness (2021, background): “The shreds of evidence in [Yang et al. , 2020b; Stutz et al. , 2019] show that adversarial training might not be the optimal solution for obtaining models with desirable robustness and accuracy.”
Globally-Robust Neural Networks (2021, methodology): “Meanwhile, a large volume of prior work applies (cid:96) ∞ -speciﬁc certiﬁcation strategies that proven effective for (cid:96) ∞ certiﬁcation (Zhang et al., 2020; Balunovic & Vechev, 2020; Gowal et al., 2019).”
Do Wider Neural Networks Really Help Adversarial Robustness? (2020, background): “[68] proved that both accuracy and robustness are achievable through locally Lipschitz functions with separated data, and the gap between theory and practice is due to either failure to impose local Lipschitzness or insufficient generalization.”