BN-invariant sharpness regularizes the training model to better generalization
This work is significant for researchers and practitioners working with neural networks, especially those employing Batch Normalization, by providing a more consistent and effective method for sharpness-based regularization to improve model generalization.
The paper addresses the inconsistency of traditional sharpness measures for neural networks with Batch Normalization (BN) layers by proposing BN-Sharpness, a scale-invariant measure. This new measure is then used to regularize model training, leading to improved generalization compared to vanilla SGD across various experimental settings.
It is arguably believed that flatter minima can generalize better. However, it has been pointed out that the usual definitions of sharpness, which consider either the maxima or the integral of loss over a $δ$ ball of parameters around minima, cannot give consistent measurement for scale invariant neural networks, e.g., networks with batch normalization layer. In this paper, we first propose a measure of sharpness, BN-Sharpness, which gives consistent value for equivalent networks under BN. It achieves the property of scale invariance by connecting the integral diameter with the scale of parameter. Then we present a computation-efficient way to calculate the BN-sharpness approximately i.e., one dimensional integral along the "sharpest" direction. Furthermore, we use the BN-sharpness to regularize the training and design an algorithm to minimize the new regularized objective. Our algorithm achieves considerably better performance than vanilla SGD over various experiment settings.