Mixture Prototype Flow Matching for Open-Set Supervised Anomaly Detection

2026-05-04

AI Summary

This work addresses the limitation of unimodal Gaussian priors in open-set supervised anomaly detection, which struggle to capture the multimodal nature of normal data distributions and consequently yield ambiguous decision boundaries. To overcome this, we propose a Mixture Prototype Flow Matching framework that constructs a continuous transformation from the normal feature distribution to a structured Gaussian mixture prototype space. Within this framework, the velocity field is modeled using a multimodal Gaussian mixture prior, and a mutual information maximization regularizer is introduced to prevent prototype collapse and enhance separability between normal and anomalous samples. By integrating flow matching, Gaussian mixture modeling, and deep feature learning, our method achieves state-of-the-art performance across multiple benchmark datasets under both single- and multi-anomaly settings.

Abstract

Open-set supervised anomaly detection (OSAD) aims to identify unseen anomalies using limited anomalous supervision. However, existing prototype-based methods typically model normal data via a unimodal Gaussian prior, failing to capture inherent multi-modality and resulting in blurred decision boundaries. To address this, we propose Mixture Prototype Flow Matching (MPFM), a framework that learns a continuous transformation from normal feature distributions to a structured Gaussian mixture prototype space. Departing from traditional flow-based approaches that rely on a single velocity vector, MPFM explicitly models the velocity field as a Gaussian mixture prior where each component corresponds to a distinct normal class. This design facilitates mode-aware and semantically coherent distribution transport. Furthermore, we introduce a Mutual Information Maximization Regularizer (MIMR) to prevent prototype collapse and maximize normal-anomaly separability. Extensive experiments demonstrate that MPFM achieves state-of-the-art performance across diverse benchmarks under both single- and multi-anomaly settings.

Problem

• open-set supervised anomaly detection
• prototype-based methods
• multi-modality
• decision boundaries
• anomaly detection

Innovation

• Mixture Prototype Flow Matching
• Gaussian Mixture Prior
• Open-Set Supervised Anomaly Detection
• Mutual Information Maximization
• Mode-Aware Distribution Transport