Enhancing Sharpness-Aware Minimization by Learning Perturbation Radius

TL;DR

This paper introduces LETS, a bilevel optimization framework that learns the optimal perturbation radius for sharpness-aware minimization, enhancing model generalization across tasks.

Contribution

It proposes a novel bilevel optimization approach to automatically learn the perturbation radius in SAM, adaptable to any SAM variant.

Findings

LETS improves generalization performance on multiple datasets.

The learned perturbation radius outperforms fixed radii in experiments.

LETS is effective across different architectures and tasks.

Abstract

Sharpness-aware minimization (SAM) is to improve model generalization by searching for flat minima in the loss landscape. The SAM update consists of one step for computing the perturbation and the other for computing the update gradient. Within the two steps, the choice of the perturbation radius is crucial to the performance of SAM, but finding an appropriate perturbation radius is challenging. In this paper, we propose a bilevel optimization framework called LEarning the perTurbation radiuS (LETS) to learn the perturbation radius for sharpness-aware minimization algorithms. Specifically, in the proposed LETS method, the upper-level problem aims at seeking a good perturbation radius by minimizing the squared generalization gap between the training and validation losses, while the lower-level problem is the SAM optimization problem. Moreover, the LETS method can be combined with any variant of SAM. Experimental results on various architectures and benchmark datasets in computer vision and natural language processing demonstrate the effectiveness of the proposed LETS method in improving the performance of SAM.