Score Combining for Contrastive OOD Detection

2025-01-21

AI Summary

Unsupervised out-of-distribution (OOD) detection remains challenging in scenarios where only in-distribution training data is available, with no access to OOD samples or labels. Method: This paper introduces a contrastive score fusion framework grounded in the Generalized Likelihood Ratio Test (GLRT), the first to integrate classical statistical hypothesis testing into contrastive learning for OOD detection. Without requiring prior knowledge or additional annotations, it constructs a robust one-class discriminant statistic by fusing multi-view contrastive representation scores. Results: Evaluated on standard benchmarks—including CIFAR-10, SVHN, LSUN, ImageNet, and CIFAR-100—as well as the CIFAR-10 leave-one-out setting, the method consistently outperforms state-of-the-art approaches such as CSI, SupCSI, and mainstream p-value combination methods (Fisher, Bonferroni, Simes) across AUROC and other metrics. It achieves significant improvements in both OOD detection accuracy and generalization capability.

Technology Category

• Machine Learning: Calibration & Uncertainty Quantification
• Computer Vision: Learning & Optimization for CV
• Data Mining & Knowledge Management: Anomaly/Outlier Detection

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Web Mining and Content Analysis: Robustness and generalizability of Web mining methods
• Economics, Online Markets and Human Computation: Data quality aspects of human-annotated datasets

Abstract

In out-of-distribution (OOD) detection, one is asked to classify whether a test sample comes from a known inlier distribution or not. We focus on the case where the inlier distribution is defined by a training dataset and there exists no additional knowledge about the novelties that one is likely to encounter. This problem is also referred to as novelty detection, one-class classification, and unsupervised anomaly detection. The current literature suggests that contrastive learning techniques are state-of-the-art for OOD detection. We aim to improve on those techniques by combining/ensembling their scores using the framework of null hypothesis testing and, in particular, a novel generalized likelihood ratio test (GLRT). We demonstrate that our proposed GLRT-based technique outperforms the state-of-the-art CSI and SupCSI techniques from Tack et al. 2020 in dataset-vs-dataset experiments with CIFAR-10, SVHN, LSUN, ImageNet, and CIFAR-100, as well as leave-one-class-out experiments with CIFAR-10. We also demonstrate that our GLRT outperforms the score-combining methods of Fisher, Bonferroni, Simes, Benjamini-Hochwald, and Stouffer in our application.

Problem

• Novelty Detection
• Single-Class Classification
• Training Data Accuracy

Innovation

• Contrastive Learning
• Statistical Tool
• Novelty Detection