EPA: Neural Collapse Inspired Robust Out-of-Distribution Detector

Fuente: arXiv

Gespeichert in:

Bibliographische Detailangaben
Hauptverfasser:	Zhang, Jiawei, Chen, Yufan, Jin, Cheng, Zhu, Lei, Gu, Yuantao
Format:	Preprint
Veröffentlicht:	2024
Schlagworte:	Machine Learning Cryptography and Security
Online-Zugang:	Acceder al recurso
Tags:	Tag hinzufügen Keine Tags, Fügen Sie den ersten Tag hinzu!

_version_	1866929196684541952
author	Zhang, Jiawei Chen, Yufan Jin, Cheng Zhu, Lei Gu, Yuantao
author_facet	Zhang, Jiawei Chen, Yufan Jin, Cheng Zhu, Lei Gu, Yuantao
contents	Out-of-distribution (OOD) detection plays a crucial role in ensuring the security of neural networks. Existing works have leveraged the fact that In-distribution (ID) samples form a subspace in the feature space, achieving state-of-the-art (SOTA) performance. However, the comprehensive characteristics of the ID subspace still leave under-explored. Recently, the discovery of Neural Collapse ($\mathcal{NC}$) sheds light on novel properties of the ID subspace. Leveraging insight from $\mathcal{NC}$, we observe that the Principal Angle between the features and the ID feature subspace forms a superior representation for measuring the likelihood of OOD. Building upon this observation, we propose a novel $\mathcal{NC}$-inspired OOD scoring function, named Entropy-enhanced Principal Angle (EPA), which integrates both the global characteristic of the ID subspace and its inner property. We experimentally compare EPA with various SOTA approaches, validating its superior performance and robustness across different network architectures and OOD datasets.
format	Preprint
id	arxiv_https___arxiv_org_abs_2401_01710
institution	arXiv
publishDate	2024
record_format	arxiv
spellingShingle	EPA: Neural Collapse Inspired Robust Out-of-Distribution Detector Zhang, Jiawei Chen, Yufan Jin, Cheng Zhu, Lei Gu, Yuantao Machine Learning Cryptography and Security Out-of-distribution (OOD) detection plays a crucial role in ensuring the security of neural networks. Existing works have leveraged the fact that In-distribution (ID) samples form a subspace in the feature space, achieving state-of-the-art (SOTA) performance. However, the comprehensive characteristics of the ID subspace still leave under-explored. Recently, the discovery of Neural Collapse ($\mathcal{NC}$) sheds light on novel properties of the ID subspace. Leveraging insight from $\mathcal{NC}$, we observe that the Principal Angle between the features and the ID feature subspace forms a superior representation for measuring the likelihood of OOD. Building upon this observation, we propose a novel $\mathcal{NC}$-inspired OOD scoring function, named Entropy-enhanced Principal Angle (EPA), which integrates both the global characteristic of the ID subspace and its inner property. We experimentally compare EPA with various SOTA approaches, validating its superior performance and robustness across different network architectures and OOD datasets.
title	EPA: Neural Collapse Inspired Robust Out-of-Distribution Detector
topic	Machine Learning Cryptography and Security
url	https://arxiv.org/abs/2401.01710