Look Around and Find Out: OOD Detection with Relative Angles

2024-10-06

arXiv.org

AI Summary

To address the unreliability of deep learning models in rejecting out-of-distribution (OOD) inputs under real-world deployment, this paper proposes a novel OOD detection paradigm based on relative angular measurement: treating the in-distribution (ID) feature mean as the origin, it computes the angle between a sample’s feature vector and the estimated decision boundary direction for discrimination. This is the first approach to introduce an ID-mean-driven angular metric, breaking from conventional distance-based paradigms while ensuring scale invariance and additive compatibility across model scores. The method integrates angular geometric modeling, decision boundary direction estimation, ID statistical modeling, and supports feature-space regularization. Evaluated on CIFAR-10 and ImageNet, it achieves state-of-the-art FPR95 reductions of 0.88% and 7.74%, respectively, significantly enhancing robustness and generalization in OOD detection.

Technology Category

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

Application Category

• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Security and Privacy: Large-scale security measurements
• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation

Abstract

Deep learning systems deployed in real-world applications often encounter data that is different from their in-distribution (ID). A reliable system should ideally abstain from making decisions in this out-of-distribution (OOD) setting. Existing state-of-the-art methods primarily focus on feature distances, such as k-th nearest neighbors and distances to decision boundaries, either overlooking or ineffectively using in-distribution statistics. In this work, we propose a novel angle-based metric for OOD detection that is computed relative to the in-distribution structure. We demonstrate that the angles between feature representations and decision boundaries, viewed from the mean of in-distribution features, serve as an effective discriminative factor between ID and OOD data. Our method achieves state-of-the-art performance on CIFAR-10 and ImageNet benchmarks, reducing FPR95 by 0.88% and 7.74% respectively. Our score function is compatible with existing feature space regularization techniques, enhancing performance. Additionally, its scale-invariance property enables creating an ensemble of models for OOD detection via simple score summation.

Problem

• Detecting out-of-distribution data in deep learning models
• Using relative angles to ID structure for OOD detection
• Improving reliability by abstaining from uncertain OOD decisions

Innovation

• Uses relative angles for OOD detection
• Leverages in-distribution statistics and boundaries
• Applies scale-invariant score for ensemble strategy