CADRef: Robust Out-of-Distribution Detection via Class-Aware Decoupled Relative Feature Leveraging

2025-03-01

AI Summary

Deep neural networks (DNNs) often exhibit overconfidence on out-of-distribution (OOD) inputs, compromising their safe deployment. To address this, we propose the Class-Aware Relative Feature (CARF) framework—a post-hoc OOD detection method that requires no architectural modification or model retraining. CARF introduces class-center-guided relative feature error as a novel OOD discriminant and innovatively decouples features based on sign alignment between relative features and classification weights, thereby enhancing intra-class compactness and inter-class separability. Evaluated across standard benchmarks—including CIFAR-10/100 and SVHN—and state-of-the-art architectures such as ResNet and ViT, CARF consistently outperforms existing methods. It achieves up to a 3.27% absolute improvement in AUROC and reduces FPR95 by up to 6.32%, establishing new state-of-the-art performance in post-hoc OOD detection.

Technology Category

• Computer Vision: Adversarial Attacks & Robustness
• Machine Learning: Calibration & Uncertainty Quantification
• 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
• Responsible Web: Human-perceived consequences of algorithmic deployment on the web
• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs

Abstract

Deep neural networks (DNNs) have been widely criticized for their overconfidence when dealing with out-of-distribution (OOD) samples, highlighting the critical need for effective OOD detection to ensure the safe deployment of DNNs in real-world settings. Existing post-hoc OOD detection methods primarily enhance the discriminative power of logit-based approaches by reshaping sample features, yet they often neglect critical information inherent in the features themselves. In this paper, we propose the Class-Aware Relative Feature-based method (CARef), which utilizes the error between a sample's feature and its class-aware average feature as a discriminative criterion. To further refine this approach, we introduce the Class-Aware Decoupled Relative Feature-based method (CADRef), which decouples sample features based on the alignment of signs between the relative feature and corresponding model weights, enhancing the discriminative capabilities of CARef. Extensive experimental results across multiple datasets and models demonstrate that both proposed methods exhibit effectiveness and robustness in OOD detection compared to state-of-the-art methods. Specifically, our two methods outperform the best baseline by 2.82% and 3.27% in AUROC, with improvements of 4.03% and 6.32% in FPR95, respectively.

Problem

• Detects out-of-distribution samples in deep neural networks.
• Improves OOD detection by leveraging class-aware relative features.
• Enhances robustness and accuracy over existing OOD detection methods.

Innovation

• Class-Aware Relative Feature-based method (CARef)
• Decouples sample features via sign alignment
• Enhances OOD detection robustness and effectiveness