Selected work

Representative Papers

Anomaly Detection with Adaptive and Aggressive Rejection for Contaminated Training Data

Nov 26, 2025

In anomaly detection, training data are often contaminated by anomalous samples, yet conventional methods rely on a predefined contamination rate—rendering them ill-suited to real-world scenarios where the contamination level is both unknown and variable, especially under significant overlap between normal and anomalous distributions. To address this, we propose an Adaptive Aggressive Anomaly Rejection (AAAR) framework that synergistically integrates hard rejection—via dynamically refined z-score thresholds—with soft rejection—based on probabilistic modeling using Gaussian Mixture Models—and employs an adaptive threshold learning mechanism for precise identification and removal of contaminated samples. Crucially, AAAR requires no prior knowledge of contamination rate. Evaluated on 2 image and 30 tabular benchmark datasets, it achieves an average AUROC improvement of 0.041 over state-of-the-art baselines, significantly enhancing model robustness, detection accuracy, and cross-domain generalization capability.

0 citationsRead paper

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

Jun 30, 2025IEEE International Joint Conference on Neural Network

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.

0 citationsRead paper

Recent publications

Latest Papers

Anomaly Detection with Adaptive and Aggressive Rejection for Contaminated Training Data

Nov 26, 2025

In anomaly detection, training data are often contaminated by anomalous samples, yet conventional methods rely on a predefined contamination rate—rendering them ill-suited to real-world scenarios where the contamination level is both unknown and variable, especially under significant overlap between normal and anomalous distributions. To address this, we propose an Adaptive Aggressive Anomaly Rejection (AAAR) framework that synergistically integrates hard rejection—via dynamically refined z-score thresholds—with soft rejection—based on probabilistic modeling using Gaussian Mixture Models—and employs an adaptive threshold learning mechanism for precise identification and removal of contaminated samples. Crucially, AAAR requires no prior knowledge of contamination rate. Evaluated on 2 image and 30 tabular benchmark datasets, it achieves an average AUROC improvement of 0.041 over state-of-the-art baselines, significantly enhancing model robustness, detection accuracy, and cross-domain generalization capability.

0 citationsRead paper

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

Jun 30, 2025IEEE International Joint Conference on Neural Network

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.

0 citationsRead paper