Elvis Dohmatob (Facebook AI Research)
Title: Tradeoffs between test error and robustness in different learning regimes for two-layer neural networks
Abstract: Neural networks are known to be highly sensitive to adversarial examples. These may arise due to different factors, such as random initialization, or spurious correlations in the learning problem. To better understand these factors, we provide a precise study of the adversarial robustness in different scenarios, from initialization to the end of training in different regimes, as well as intermediate scenarios, where initialization still plays a role due to “lazy” training. We consider over- parameterized networks in high dimensions with quadratic targets and infinite samples. Our analysis allows us to identify new tradeoffs between test error and robustness, whereby robustness can only get worse when test error improves, and vice versa.
Joint work with Alberto Bietti. Paper link: https://arxiv.org/abs/2203.11864
If you would like to attend, please email [email protected] for the Zoom details.