An Evidence-Based Post-Hoc Adjustment Framework for Anomaly Detection Under Data Contamination

2025-10-24

AI Summary

Real-world unsupervised anomaly detection often suffers severe performance degradation when training data contain unlabeled or mislabeled anomalies. Existing robust methods typically require access to training procedures, strong data priors, or assumptions about anomaly prevalence—limiting practical applicability. To address this, we propose EPHAD: the first training-agnostic, test-time post-processing adaptive framework for unsupervised anomaly detection. EPHAD dynamically refines detector outputs by fusing heterogeneous external evidence—including multimodal foundation models (e.g., CLIP) and classical detectors (e.g., Latent Outlier Factor)—without modifying the original model architecture or training pipeline. Its core innovations are a principled evidence integration mechanism and a Bayesian posterior adjustment strategy. Evaluated on 8 vision, 26 tabular, and 1 industrial dataset, EPHAD consistently enhances robustness and generalization across diverse detectors under varying contamination levels, demonstrating strong universality and plug-and-play compatibility.

Technology Category

• Machine Learning: Adversarial Learning & Robustness
• Data Mining & Knowledge Management: Anomaly/Outlier Detection
• Computer Vision: Adversarial Attacks & Robustness

Application Category

• Economics, Online Markets and Human Computation: Humans versus LLMs for data annotation and labeling
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Unsupervised anomaly detection (AD) methods typically assume clean training data, yet real-world datasets often contain undetected or mislabeled anomalies, leading to significant performance degradation. Existing solutions require access to the training pipelines, data or prior knowledge of the proportions of anomalies in the data, limiting their real-world applicability. To address this challenge, we propose EPHAD, a simple yet effective test-time adaptation framework that updates the outputs of AD models trained on contaminated datasets using evidence gathered at test time. Our approach integrates the prior knowledge captured by the AD model trained on contaminated datasets with evidence derived from multimodal foundation models like Contrastive Language-Image Pre-training (CLIP), classical AD methods like the Latent Outlier Factor or domain-specific knowledge. We illustrate the intuition behind EPHAD using a synthetic toy example and validate its effectiveness through comprehensive experiments across eight visual AD datasets, twenty-six tabular AD datasets, and a real-world industrial AD dataset. Additionally, we conduct an ablation study to analyse hyperparameter influence and robustness to varying contamination levels, demonstrating the versatility and robustness of EPHAD across diverse AD models and evidence pairs. To ensure reproducibility, our code is publicly available at https://github.com/sukanyapatra1997/EPHAD.

Problem

• Addressing performance degradation in unsupervised anomaly detection under data contamination
• Proposing test-time adaptation framework without requiring training pipeline access
• Integrating prior knowledge with multimodal evidence for robust anomaly detection

Innovation

• Test-time adaptation framework for contaminated data
• Integrates prior knowledge with multimodal foundation models
• Validated across diverse visual and tabular datasets