RCR-AF: Enhancing Model Generalization via Rademacher Complexity Reduction Activation Function

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Yu, Yunrui, Wang, Kafeng, Su, Hang, Zhu, Jun
Format: Preprint
Veröffentlicht: 2025
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866911083658215424
author Yu, Yunrui
Wang, Kafeng
Su, Hang
Zhu, Jun
author_facet Yu, Yunrui
Wang, Kafeng
Su, Hang
Zhu, Jun
contents Despite their widespread success, deep neural networks remain critically vulnerable to adversarial attacks, posing significant risks in safety-sensitive applications. This paper investigates activation functions as a crucial yet underexplored component for enhancing model robustness. We propose a Rademacher Complexity Reduction Activation Function (RCR-AF), a novel activation function designed to improve both generalization and adversarial resilience. RCR-AF uniquely combines the advantages of GELU (including smoothness, gradient stability, and negative information retention) with ReLU's desirable monotonicity, while simultaneously controlling both model sparsity and capacity through built-in clipping mechanisms governed by two hyperparameters, $α$ and $γ$. Our theoretical analysis, grounded in Rademacher complexity, demonstrates that these parameters directly modulate the model's Rademacher complexity, offering a principled approach to enhance robustness. Comprehensive empirical evaluations show that RCR-AF consistently outperforms widely-used alternatives (ReLU, GELU, and Swish) in both clean accuracy under standard training and in adversarial robustness within adversarial training paradigms.
format Preprint
id arxiv_https___arxiv_org_abs_2507_22446
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle RCR-AF: Enhancing Model Generalization via Rademacher Complexity Reduction Activation Function
Yu, Yunrui
Wang, Kafeng
Su, Hang
Zhu, Jun
Machine Learning
Artificial Intelligence
Computer Vision and Pattern Recognition
Despite their widespread success, deep neural networks remain critically vulnerable to adversarial attacks, posing significant risks in safety-sensitive applications. This paper investigates activation functions as a crucial yet underexplored component for enhancing model robustness. We propose a Rademacher Complexity Reduction Activation Function (RCR-AF), a novel activation function designed to improve both generalization and adversarial resilience. RCR-AF uniquely combines the advantages of GELU (including smoothness, gradient stability, and negative information retention) with ReLU's desirable monotonicity, while simultaneously controlling both model sparsity and capacity through built-in clipping mechanisms governed by two hyperparameters, $α$ and $γ$. Our theoretical analysis, grounded in Rademacher complexity, demonstrates that these parameters directly modulate the model's Rademacher complexity, offering a principled approach to enhance robustness. Comprehensive empirical evaluations show that RCR-AF consistently outperforms widely-used alternatives (ReLU, GELU, and Swish) in both clean accuracy under standard training and in adversarial robustness within adversarial training paradigms.
title RCR-AF: Enhancing Model Generalization via Rademacher Complexity Reduction Activation Function
topic Machine Learning
Artificial Intelligence
Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2507.22446