Logits-Based Finetuning

2025-05-30

AI Summary

This work addresses out-of-distribution (OOD) detection by learning in-distribution (ID) representations that effectively discriminate between ID and OOD samples. To mitigate feature shortcuts commonly introduced by discriminative methods, we propose leveraging reconstruction-based pretraining—specifically masked image modeling (MIM)—as a general-purpose prior to steer representation learning toward the intrinsic structural properties of ID data. Based on this insight, we introduce MOOD: a framework that performs reconstruction-driven representation learning and logits-level fine-tuning using ID data only—without requiring any OOD samples. We provide the first systematic evidence that reconstruction tasks serve as an efficient and broadly applicable prior for OOD detection. Experiments demonstrate that MOOD improves performance by 5.7%, 3.0%, and 2.1% on single-class, multi-class, and near-distribution OOD detection benchmarks, respectively, and consistently surpasses state-of-the-art 10-shot anomaly exposure methods that rely on OOD samples.

Technology Category

• Machine Learning: Representation Learning
• Computer Vision: Representation Learning for Vision
• Data Mining & Knowledge Management: Anomaly/Outlier Detection

Application Category

• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

The core of out-of-distribution (OOD) detection is to learn the in-distribution (ID) representation, which is distinguishable from OOD samples. Previous work applied recognition-based methods to learn the ID features, which tend to learn shortcuts instead of comprehensive representations. In this work, we find surprisingly that simply using reconstruction-based methods could boost the performance of OOD detection significantly. We deeply explore the main contributors of OOD detection and find that reconstruction-based pretext tasks have the potential to provide a generally applicable and efficacious prior, which benefits the model in learning intrinsic data distributions of the ID dataset. Specifically, we take Masked Image Modeling as a pretext task for our OOD detection framework (MOOD). Without bells and whistles, MOOD outperforms previous SOTA of one-class OOD detection by 5.7%, multi-class OOD detection by 3.0%, and near-distribution OOD detection by 2.1%. It even defeats the 10-shot-per-class outlier exposure OOD detection, although we do not include any OOD samples for our detection. Codes are available at https://github.com/JulietLJY/MOOD.

Problem

• Enhancing OOD detection via reconstruction-based methods
• Learning intrinsic ID data distributions effectively
• Outperforming SOTA without using OOD samples

Innovation

• Uses reconstruction-based methods for OOD detection
• Employs Masked Image Modeling as pretext task
• Achieves SOTA without OOD samples