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

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Lee, Jungi, Kim, Jungkwon, Zhang, Chi, Kim, Sangmin, Yoo, Kwangsun, Byun, Seok-Joo
Natura: Preprint
Pubblicazione: 2026
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866911356255469568
author Lee, Jungi
Kim, Jungkwon
Zhang, Chi
Kim, Sangmin
Yoo, Kwangsun
Byun, Seok-Joo
author_facet Lee, Jungi
Kim, Jungkwon
Zhang, Chi
Kim, Sangmin
Yoo, Kwangsun
Byun, Seok-Joo
contents 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.
format Preprint
id arxiv_https___arxiv_org_abs_2601_02440
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Mitigating Long-Tailed Anomaly Score Distributions with Importance-Weighted Loss
Lee, Jungi
Kim, Jungkwon
Zhang, Chi
Kim, Sangmin
Yoo, Kwangsun
Byun, Seok-Joo
Machine Learning
Artificial Intelligence
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.
title Mitigating Long-Tailed Anomaly Score Distributions with Importance-Weighted Loss
topic Machine Learning
Artificial Intelligence
url https://arxiv.org/abs/2601.02440