ExGrad: Beyond Suppression - Explicit Clean-Gradient Subtraction in Adversarial Training

Paper recorded by Signals 4 on 2026-10-21 in cs.LG. Abstract reproduced from arXiv; link to the original below.

Published 2026-10-21 on arXiv · recorded by Signals 4 on 2026-10-22

Category: cs.LG · 机器学习 · first seen 2026-10-22

Abstract

Adversarial training derives updates from surrogate losses, but source-loss descent does not directly characterize an update's effect on adversarial margins. We introduce ExGrad (extrapolative gradient recomposition), which constructs clean-input, adversarial-input, and adversarial-weight-perturbed (AWP) gradients. Virtual steps measure their margin responses, and a globally solvable three-variabl

Read on arXiv →