Lookbehind-SAM: k steps back, 1 step forward

TL;DR

Lookbehind-SAM introduces multiple ascent steps behind the current point to better identify worst-case perturbations, improving generalization, robustness, and lifelong learning performance.

Contribution

It proposes a novel Lookbehind method that enhances SAM by performing multiple ascent steps behind, inspired by Lookahead, to improve loss-sharpness trade-offs.

Findings

Enhanced generalization performance across tasks

Increased robustness against noisy weights

Improved learning and reduced catastrophic forgetting

Abstract

Sharpness-aware minimization (SAM) methods have gained increasing popularity by formulating the problem of minimizing both loss value and loss sharpness as a minimax objective. In this work, we increase the efficiency of the maximization and minimization parts of SAM's objective to achieve a better loss-sharpness trade-off. By taking inspiration from the Lookahead optimizer, which uses multiple descent steps ahead, we propose Lookbehind, which performs multiple ascent steps behind to enhance the maximization step of SAM and find a worst-case perturbation with higher loss. Then, to mitigate the variance in the descent step arising from the gathered gradients across the multiple ascent steps, we employ linear interpolation to refine the minimization step. Lookbehind leads to a myriad of benefits across a variety of tasks. Particularly, we show increased generalization performance, greater robustness against noisy weights, as well as improved learning and less catastrophic forgetting in lifelong learning settings. Our code is available at https://github.com/chandar-lab/Lookbehind-SAM.