Dissecting Mahalanobis: How Feature Geometry and Normalization Shape OOD Detection

2025-10-16

AI Summary

Mahalanobis distance is widely used for out-of-distribution (OOD) detection, yet its performance is highly sensitive to the geometric structure of deep features and normalization schemes—mechanisms that remain poorly understood. Method: This work systematically establishes intrinsic connections between feature-space geometry—including spectral properties and intrinsic dimensionality—and OOD detection performance. We demonstrate that standard ℓ₂ normalization is suboptimal for Mahalanobis-based detection and propose a tunable radial scaling ℓ₂ normalization that explicitly controls feature magnitude distributions to better satisfy the underlying assumptions of Mahalanobis distance. Contribution/Results: Extensive experiments across multiple vision foundation models and benchmark datasets show that our method significantly improves OOD detection accuracy, overcoming key limitations of conventional Mahalanobis approaches. Moreover, the derived geometry–performance mapping provides theoretical foundations for OOD detection, enhancing both model robustness and interpretability.

Technology Category

• Machine Learning: Other Foundations of Machine Learning
• Computer Vision: Other Foundations of Computer Vision
• Data Mining & Knowledge Management: Anomaly/Outlier Detection

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Security and Privacy: Large-scale security measurements
• Web Mining and Content Analysis: Normalization, clustering, classification, and summarization of Web text

Abstract

Out-of-distribution (OOD) detection is critical for the reliable deployment of deep learning models. hile Mahalanobis distance methods are widely used, the impact of representation geometry and normalization on their performance is not fully understood, which may limit their downstream application. To address this gap, we conducted a comprehensive empirical study across diverse image foundation models, datasets, and distance normalization schemes. First, our analysis shows that Mahalanobis-based methods aren't universally reliable. Second, we define the ideal geometry for data representations and demonstrate that spectral and intrinsic-dimensionality metrics can accurately predict a model's OOD performance. Finally, we analyze how normalization impacts OOD performance. Building upon these studies, we propose radially scaled $ell_2$ normalization, a method that generalizes the standard $ell_2$ normalization recently applied to Mahalanobis-based OOD detection. Our approach introduces a tunable parameter to directly control the radial geometry of the feature space, systematically contracting or expanding representations to significantly improve OOD detection performance. By bridging the gap between representation geometry, normalization, and OOD performance, our findings offer new insights into the design of more effective and reliable deep learning models.

Problem

• Analyzing feature geometry impact on Mahalanobis OOD detection
• Investigating normalization effects on out-of-distribution detection performance
• Developing improved normalization method for reliable OOD detection

Innovation

• Radially scaled L2 normalization controls feature geometry
• Spectral metrics predict model out-of-distribution performance
• Tunable parameter systematically contracts or expands representations