Skip to yearly menu bar Skip to main content


Poster

Normalization Layers Are All That Sharpness-Aware Minimization Needs

Maximilian Mueller · Tiffany Vlaar · David Rolnick · Matthias Hein

Great Hall & Hall B1+B2 (level 1) #517

Abstract:

Sharpness-aware minimization (SAM) was proposed to reduce sharpness of minima and has been shown to enhance generalization performance in various settings. In this work we show that perturbing only the affine normalization parameters (typically comprising 0.1% of the total parameters) in the adversarial step of SAM can outperform perturbing all of the parameters. This finding generalizesto different SAM variants and both ResNet (Batch Normalization) and Vision Transformer (Layer Normalization) architectures. We consider alternative sparse perturbation approaches and find that these do not achieve similar performance enhancement at such extreme sparsity levels, showing that this behaviour is unique to the normalization layers. Although our findings reaffirm the effectivenessof SAM in improving generalization performance, they cast doubt on whether this is solely caused by reduced sharpness.

Chat is not available.