Saved in:
Bibliographic Details
Main Authors: Zou, Xiandong, Li, Jia, Yuan, Xiaotong, Zhou, Pan
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2510.25262
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866912858856488960
author Zou, Xiandong
Li, Jia
Yuan, Xiaotong
Zhou, Pan
author_facet Zou, Xiandong
Li, Jia
Yuan, Xiaotong
Zhou, Pan
contents Normalization is fundamental to deep learning, but existing approaches such as BatchNorm, LayerNorm, and RMSNorm are variance-centric by enforcing zero mean and unit variance, stabilizing training without controlling how representations capture task-relevant information. We propose IB-Inspired Normalization (IBNorm), a simple yet powerful family of methods grounded in the Information Bottleneck principle. IBNorm introduces bounded compression operations that encourage embeddings to preserve predictive information while suppressing nuisance variability, yielding more informative representations while retaining the stability and compatibility of standard normalization. Theoretically, we prove that IBNorm achieves a higher IB value and tighter generalization bounds than variance-centric methods. Empirically, IBNorm consistently outperforms BatchNorm, LayerNorm, and RMSNorm across large-scale language models (LLaMA, GPT-2) and vision models (ResNet, ViT), with mutual information analysis confirming superior information bottleneck behavior. Code will be released publicly.
format Preprint
id arxiv_https___arxiv_org_abs_2510_25262
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle IBNorm: Information-Bottleneck Inspired Normalization for Representation Learning
Zou, Xiandong
Li, Jia
Yuan, Xiaotong
Zhou, Pan
Machine Learning
Artificial Intelligence
Normalization is fundamental to deep learning, but existing approaches such as BatchNorm, LayerNorm, and RMSNorm are variance-centric by enforcing zero mean and unit variance, stabilizing training without controlling how representations capture task-relevant information. We propose IB-Inspired Normalization (IBNorm), a simple yet powerful family of methods grounded in the Information Bottleneck principle. IBNorm introduces bounded compression operations that encourage embeddings to preserve predictive information while suppressing nuisance variability, yielding more informative representations while retaining the stability and compatibility of standard normalization. Theoretically, we prove that IBNorm achieves a higher IB value and tighter generalization bounds than variance-centric methods. Empirically, IBNorm consistently outperforms BatchNorm, LayerNorm, and RMSNorm across large-scale language models (LLaMA, GPT-2) and vision models (ResNet, ViT), with mutual information analysis confirming superior information bottleneck behavior. Code will be released publicly.
title IBNorm: Information-Bottleneck Inspired Normalization for Representation Learning
topic Machine Learning
Artificial Intelligence
url https://arxiv.org/abs/2510.25262