Denoising without Diffusion: Fixed-Noise Denoiser Anomaly Detection in Tabular Data

While diffusion models have advanced anomaly detection, their reliance on multi-step noise schedules introduces significant computational complexity. In this paper, we demonstrate that the generative capability of diffusion is not required for tabular anomaly detection. We revisit core principles of denoising without targeting data generation and present a deep-learning approach that streamlines these objectives into a fixed-noise formulation. Unlike denoising autoencoders that rely on reconstruction error, our method utilizes a preconditioning with an explicit linear reference channel. We train a self-supervised fixednoise denoising predictor and derive an anomaly score from the expected deviation under repeated perturbations, yielding a stability proxy rather than merely measuring distance to the data manifold. On the well-established ADBench benchmark, our method achieves state-of-the-art performance with improvements over existing baselines of 1.22% in AUCROC and 1.13% in AUCPR, the most informative and threshold-independent metrics. Our approach emphasizes structural simplicity and efficiency, demonstrating that a singlestep, stability-based objective outperforms complex generative schedules.

Recently, diffusion models (Ho et al., 2020; Dhariwal & Nichol, 2021) and denoising-based objectives (Song & Ermon, 2019; Song et al., 2020) have emerged as powerful • The score admits a bias-stability decomposition, corresponding to a Jacobian-norm surrogate under a firstorder expansion. This decomposition probes the local instability of the learned denoising field without requiring explicit likelihood estimation. Empirically, the resulting neighborhood score outperforms reconstruction-based scoring, while a direct Jacobianonly baseline is insufficient.

• We show that timestep conditioning, schedules, and multi-step reverse processes are not necessary for oneclass tabular anomaly detection, and that a single fixednoise denoiser with an appropriate score is sufficient. This leads to a substantial simplification, reducing both computational complexity and memory requirements.

• We achieve state-of-the-art performance with consistent improvements over prior diffusion-based and classical baselines on ADBench across various architectures, outperforming prior baselines across all metrics.

Anomaly detection has been studied extensively across a wide range of application domains. In this section, we briefly review related categories. Comprehensive comparisons are presented in the following surveys, Ruff et al.