# $L\_2$BN: Enhancing Batch Normalization by Equalizing the $L\_2$ Norms of Features

Source: [$L\_2$BN: Enhancing Batch Normalization by Equalizing the $L\_2$ Norms of Features](<https://www.sota2.com/research/paper/68fa6ed21e209131cea426bd>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2025-10-26T01:07:49.028Z

Authors: Zhennan Wang, Kehan Li, Runyi Yu, Yian Zhao, Pengchong Qiao, Chang Liu, Fan Xu, Xiangyang Ji, Guoli Song, Jie Chen

Published: 2022-07-06T12:34:33.000Z

## Abstract

In this paper, we analyze batch normalization from the perspective of discriminability and find the disadvantages ignored by previous studies: the difference in $l\_2$ norms of sample features can hinder batch normalization from obtaining more distinguished inter-class features and more compact intra-class features. To address this issue, we propose a simple yet effective method to equalize the $l\_2$ norms of sample features. Concretely, we $l\_2$-normalize each sample feature before feeding them into batch normalization, and therefore the features are of the same magnitude. Since the proposed method combines the $l\_2$ normalization and batch normalization, we name our method $L\_2$BN. The $L\_2$BN can strengthen the compactness of intra-class features and enlarge the discrepancy of inter-class features. The $L\_2$BN is easy to implement and can exert its effect without any additional parameters or hyper-parameters. We evaluate the effectiveness of $L\_2$BN through extensive experiments with various models on image classification and acoustic scene classification tasks. The results demonstrate that the $L\_2$BN can boost the generalization ability of various neural network models and achieve considerable performance improvements.

## Recorded benchmark results

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

## Original sources and implementation

[Original paper](<https://arxiv.org/abs/2207.02625>)

[PDF](<https://arxiv.org/pdf/2207.02625.pdf>)
