Mitigating Long-Tailed Anomaly Score Distributions with Importance-Weighted Loss

2025-06-30

IEEE International Joint Conference on Neural Network

AI Summary

This work addresses the long-tailed distribution (LTD) of anomaly scores in industrial anomaly detection, which arises from the diversity of normal samples and leads to biased model training and degraded detection performance for underrepresented normal patterns. To mitigate this issue, the authors propose a novel importance-weighted loss function that requires no prior knowledge of normal sample categories. For the first time, they align the anomaly score distribution to a target Gaussian distribution via importance sampling under a fully unsupervised setting. This approach effectively alleviates the LTD problem and enhances the model’s ability to equitably represent diverse normal modes. Extensive experiments on three image benchmarks and three real-world hyperspectral datasets demonstrate an average improvement of 0.043 in detection performance, confirming the method’s effectiveness and generalizability.

Technology Category

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

Application Category

• Graph Algorithms and Modeling for the Web: Algorithms and analysis for incomplete, noisy, or partially observed Web-related graphs
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Web Mining and Content Analysis: Normalization, clustering, classification, and summarization of Web text

Abstract

Anomaly detection is crucial in industrial applications for identifying rare and unseen patterns to ensure system reliability. Traditional models, trained on a single class of normal data, struggle with real-world distributions where normal data exhibit diverse patterns, leading to class imbalance and long-tailed anomaly score distributions (LTD). This imbalance skews model training and degrades detection performance, especially for minority instances. To address this issue, we propose a novel importance-weighted loss designed specifically for anomaly detection. Compared to the previous method for LTD in classification, our method does not require prior knowledge of normal data classes. Instead, we introduce a weighted loss function that incorporates importance sampling to align the distribution of anomaly scores with a target Gaussian, ensuring a balanced representation of normal data. Extensive experiments on three benchmark image datasets and three real-world hyperspectral imaging datasets demonstrate the robustness of our approach in mitigating LTD-induced bias. Our method improves anomaly detection performance by 0.043, highlighting its effectiveness in real-world applications.

Problem

• anomaly detection
• long-tailed distribution
• class imbalance
• anomaly score distribution
• industrial applications

Innovation

• importance-weighted loss
• long-tailed anomaly score distribution
• anomaly detection
• importance sampling
• class imbalance