Enhancing OOD Detection Using Latent Diffusion

2024-06-24

AI Summary

To address pixel-level generation instability in out-of-distribution (OOD) sample synthesis using auxiliary data for OOD detection in open-world scenarios, this paper proposes the Outlier Synthesis in Latent Space (OAL) framework. OAL synthesizes high-fidelity OOD samples in the latent space of Stable Diffusion—bypassing pixel-space instabilities inherent in training. It introduces a mutual information-driven contrastive learning (MICL) module to explicitly enhance separability between in-distribution (ID) and OOD features. Additionally, a knowledge distillation mechanism preserves near-lossless ID classification accuracy while boosting OOD detection performance. Integrating latent diffusion modeling, mutual information maximization, and contrastive learning, OAL supports lightweight fine-tuning. Evaluated on CIFAR-10/100 benchmarks, OAL achieves state-of-the-art OOD detection performance using only a small number of synthesized OOD samples—significantly outperforming existing pixel-space synthesis methods.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Large Multimodal Models (LMMs)
• Data Mining & Knowledge Management: Anomaly/Outlier Detection

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Numerous Out-of-Distribution (OOD) detection algorithms have been developed to identify unknown samples or objects in real-world deployments. One line of work related to OOD detection propose utilizing auxiliary datasets to train OOD detectors, thereby enhancing the performance of OOD detection. Recently, researchers propose to leverage Stable Diffusion (SD) to generate outliers in the pixel space, which may complicate network training. To mitigate this issue, we propose an Outlier Aware Learning (OAL) framework, which synthesizes OOD training data in the latent space. This improvement enables us to train the network with only a few synthesized outliers. Besides, to regularize the model's decision boundary, we develop a mutual information-based contrastive learning module (MICL) that amplifies the distinction between In-Distribution (ID) and collected OOD features. Moreover, we develop a knowledge distillation module to prevent the degradation of ID classification accuracy when training with OOD data. Extensive experiments on CIFAR-10/100 benchmarks demonstrate the superior performance of our method.

Problem

• Improving OOD detection using latent space synthesis
• Enhancing ID-OOD distinction via contrastive learning
• Maintaining ID classification accuracy during OOD training

Innovation

• OAL framework synthesizes OOD data in latent space
• MICL module enhances ID and OOD feature distinction
• Knowledge distillation maintains ID classification accuracy