Generating Clean Samples from Extremely Low SNR Data: Noise-Robust CFM
kwangmoo_yi · x · 2026-08-05
Shares the arXiv paper Noise-Robust Conditional Flow Matching: Generating Clean Samples from Noisy Datasets.
The paper introduces NR-CFM (Noise-Robust Conditional Flow Matching), a method designed to learn and generate clean samples directly from noisy observations. In fields like scientific imaging, obtaining clean reference data is often prohibitively expensive, and training directly on noisy measurements causes generative models to reproduce corrupted data.
NR-CFM provides a closed-form clean endpoint correction for additive white Gaussian noise and learns data-driven corrections for general Gaussian corruptions with complex covariance structures. Experiments demonstrate that NR-CFM outperforms NR-GAN in most settings and remains competitive with Ambient Diffusion in high-noise regimes. It can even generate plausible samples from scientific data at signal-to-noise ratios as low as 0.001.
Related event: Noise-Robust Conditional Flow Matching Generates Clean Samples(2 posts)→
More from Research
- Do Big Lab Researchers Read Zero Papers? The ML Reproducibility Crisis — alex_peys · 2026-08-05
- MSR Paper Proposes 'Decision-Analytic Steering' for Safe LLM High-Stakes Decisions — brwilder · 2026-08-05
- ECCV 2026 Paper Introduces MegaKPT Dataset with 1.3M Instances and DINOv3-based GKDT Model — NielsRogge · 2026-08-05
- Stanford Paper Demystifies Why Vision-Language-Action Models Fail in Contact-Rich Tasks — StanfordAILab · 2026-08-05
- Abnormally Dead NeurIPS Review Period: Are Authors and Reviewers Ghosting? — RevolutionaryPea8272 · 2026-08-05
- AI Scientist Summer Workshop Announces Agenda: Reliability and Closed-Loop Discovery — nc_frey · 2026-08-05