Research map / Image generation / Fast diffusion sampling
Denoising without Diffusion: Fixed-Noise Denoiser Anomaly Detection in Tabular Data
Manuel Hirth, Lukas Koberg, Nasser Jazdi, Enkelejda Kasneci
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use fixed noise without timestep embedding instead of EDM schedule (ablation) | DenoiserAD · ADBench | AUCROC (%): 69.71 -> 83.34 | Improved |
| use fixed noise instead of noise schedule NCSN/NCSBAD (ablation) | DenoiserAD · ADBench | AUCROC (%): 81.86 -> 83.34 | Improved |
| add timestep embedding to fixed noise (ablation) | DenoiserAD · ADBench | AUCROC (%): 83.34 -> 83.11 | Hurt |
| sample noise from stable interval [0.12, 0.35] instead of fixed sigma^2 (ablation) | DenoiserAD · ADBench | AUCROC (%): 83.34 -> 81.78 | Hurt |
| add explicit linear reference channel (vs plain DAE) with stability score S2 (ablation) | DenoiserAD · ADBench | AUCROC (%): 82.81 -> 83.34 | Improved |
| use Monte-Carlo input-referenced score S2 instead of single-shot reconstruction S1 on plain DAE (ablation) | DAE · ADBench | AUCROC (%): 82.38 -> 82.81 (+0.43) | No clear effect |
| use stability score S2 instead of augmented reconstruction S4 (ablation) | DenoiserAD · ADBench | AUCROC (%): 81.33 -> 83.34 | Improved |
| use stability score S2 instead of mean-centered variance S3 (ablation) | DenoiserAD · ADBench | AUCROC (%): 75.88 -> 83.34 | Improved |
| use stability score S2 instead of Jacobian-norm score (ablation) | DenoiserAD · ADBench | AUCROC (%): 81.05 -> 83.34 | Improved |
| apply fixed-noise stability formulation to DDPM-based AD (ablation) | DDPM · ADBench | AUCROC (%): 70.73 -> 77.91 | Improved |
| apply fixed-noise stability formulation to DTE (ablation) | DTE · ADBench | AUCROC (%): 79.77 -> 80.65 | Improved |
| replace DTE backbone with ResMLP only (ablation) | DTE · ADBench | AUCROC (%): 79.77 -> 69.39 | Hurt |
Closest papers by meaning
- Interpreting and Improving Diffusion Models from an Optimization Perspective (ICML 2024 · similarity 0.87)
- Denoising Diffusion Bridge Models (ICLR 2024 · similarity 0.87)
- Your Diffusion Model is Secretly a Noise Classifier and Benefits from Contrastive Training (NeurIPS 2024 · similarity 0.87)
- Is Noise Conditioning Necessary for Denoising Generative Models? (ICML 2025 · similarity 0.87)
- Training-Free Safe Denoisers for Safe Use of Diffusion Models (NeurIPS 2025 · similarity 0.87)
- Think while You Generate: Discrete Diffusion with Planned Denoising (ICLR 2025 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
