Improved Deep Neural Network Generalization Using m-Sharpness-Aware Minimization
This work addresses the challenge of reliable optimization for over-parameterized deep learning models, but it is incremental as it builds on existing SAM methods with a thorough empirical study.
The paper tackles the problem of improving generalization in deep neural networks by evaluating m-Sharpness-Aware Minimization (mSAM), a variant that averages adversarial perturbations across mini-batch shards. The result shows that mSAM yields superior generalization performance and flatter minima compared to SAM across various tasks without significantly increasing computational costs.
Modern deep learning models are over-parameterized, where the optimization setup strongly affects the generalization performance. A key element of reliable optimization for these systems is the modification of the loss function. Sharpness-Aware Minimization (SAM) modifies the underlying loss function to guide descent methods towards flatter minima, which arguably have better generalization abilities. In this paper, we focus on a variant of SAM known as mSAM, which, during training, averages the updates generated by adversarial perturbations across several disjoint shards of a mini-batch. Recent work suggests that mSAM can outperform SAM in terms of test accuracy. However, a comprehensive empirical study of mSAM is missing from the literature -- previous results have mostly been limited to specific architectures and datasets. To that end, this paper presents a thorough empirical evaluation of mSAM on various tasks and datasets. We provide a flexible implementation of mSAM and compare the generalization performance of mSAM to the performance of SAM and vanilla training on different image classification and natural language processing tasks. We also conduct careful experiments to understand the computational cost of training with mSAM, its sensitivity to hyperparameters and its correlation with the flatness of the loss landscape. Our analysis reveals that mSAM yields superior generalization performance and flatter minima, compared to SAM, across a wide range of tasks without significantly increasing computational costs.