Probability Density from Latent Diffusion Models for Out-of-Distribution Detection

2025-08-21

AI Summary

Likelihood-based out-of-distribution (OOD) detection fails in latent diffusion models (LDMs), primarily due to limitations in pixel-space modeling—not inherent unreliability of representation-space features. Theoretically, under uniform distribution assumptions, data likelihood is the optimal OOD detector; empirically, representation space yields more reliable likelihood estimates than pixel space. Method: We propose a variational diffusion model operating in the feature space of a pretrained ResNet-18, enabling probabilistic density estimation at the representation level. Our approach is systematically evaluated on standard OOD benchmarks using the OpenOOD framework. Contribution/Results: The method substantially outperforms existing likelihood-based OOD detectors. It is the first to demonstrate that high-quality learned representations can support robust, highly discriminative likelihood-based OOD detection—establishing a new paradigm for safe deployment of generative models.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Large Multimodal Models (LMMs)
• Reasoning under Uncertainty: Relational Probabilistic Models

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Despite rapid advances in AI, safety remains the main bottleneck to deploying machine-learning systems. A critical safety component is out-of-distribution detection: given an input, decide whether it comes from the same distribution as the training data. In generative models, the most natural OOD score is the data likelihood. Actually, under the assumption of uniformly distributed OOD data, the likelihood is even the optimal OOD detector, as we show in this work. However, earlier work reported that likelihood often fails in practice, raising doubts about its usefulness. We explore whether, in practice, the representation space also suffers from the inability to learn good density estimation for OOD detection, or if it is merely a problem of the pixel space typically used in generative models. To test this, we trained a Variational Diffusion Model not on images, but on the representation space of a pre-trained ResNet-18 to assess the performance of our likelihood-based detector in comparison to state-of-the-art methods from the OpenOOD suite.

Problem

• Detecting out-of-distribution data using likelihood scores
• Evaluating representation space vs pixel space for OOD detection
• Assessing Variational Diffusion Models on pre-trained feature representations

Innovation

• Latent space diffusion for OOD detection
• ResNet-18 feature space density estimation
• Variational Diffusion Model likelihood evaluation