Diffusion-Scheduled Denoising Autoencoders for Anomaly Detection in Tabular Data

2025-08-01

AI Summary

Addressing the challenges of complex feature interactions and scarce anomalous samples in tabular anomaly detection, this paper proposes DiffCL—a denoising autoencoder framework that integrates diffusion-model noise scheduling with contrastive learning. DiffCL embeds temporally controlled diffusion-style noise injection into an encoder-decoder architecture and employs contrastive learning to enhance the representational separability between normal and anomalous samples, thereby improving discriminative capability under both semi-supervised and unsupervised settings. Extensive experiments across 57 benchmark datasets demonstrate that DiffCL achieves substantial gains: +9 percentage points (65% relative improvement) in PR-AUC and +6 percentage points (16% relative improvement) in ROC-AUC under semi-supervised evaluation, significantly outperforming state-of-the-art autoencoder- and diffusion-based baselines. These results validate DiffCL’s strong robustness and generalization across diverse anomaly distributions.

Technology Category

• Data Mining & Knowledge Management: Anomaly/Outlier Detection
• Machine Learning: Deep Generative Models & Autoencoders
• Computer Vision: Diffusion Models for Vision

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Economics, Online Markets and Human Computation: Data quality aspects of human-annotated datasets
• Web Mining and Content Analysis: Normalization, clustering, classification, and summarization of Web text

Abstract

Anomaly detection in tabular data remains challenging due to complex feature interactions and the scarcity of anomalous examples. Denoising autoencoders rely on fixed-magnitude noise, limiting adaptability to diverse data distributions. Diffusion models introduce scheduled noise and iterative denoising, but lack explicit reconstruction mappings. We propose the Diffusion-Scheduled Denoising Autoencoder (DDAE), a framework that integrates diffusion-based noise scheduling and contrastive learning into the encoding process to improve anomaly detection. We evaluated DDAE on 57 datasets from ADBench. Our method outperforms in semi-supervised settings and achieves competitive results in unsupervised settings, improving PR-AUC by up to 65% (9%) and ROC-AUC by 16% (6%) over state-of-the-art autoencoder (diffusion) model baselines. We observed that higher noise levels benefit unsupervised training, while lower noise with linear scheduling is optimal in semi-supervised settings. These findings underscore the importance of principled noise strategies in tabular anomaly detection.

Problem

• Challenges in detecting anomalies in tabular data due to complex feature interactions
• Limitations of fixed-magnitude noise in denoising autoencoders for diverse data
• Lack of explicit reconstruction mappings in diffusion models for anomaly detection

Innovation

• Integrates diffusion-based noise scheduling
• Uses contrastive learning in encoding
• Optimizes noise levels for training