Research map / Probabilistic and scientific modeling / Variational and generative sampling
Adapting Noise to Data: Generative Flows from Learned 1D Processes
Jannis Chemseddine, Gregor Kornhardt, Richard Duong, Gabriele Steidl
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 |
|---|---|---|---|
| learn per-coordinate quantile noise instead of Gaussian noise | OT flow matching · HRRR-mini precipitation | extreme event magnitude error: 0.2455 -> 0.0634 | Improved |
| learn per-coordinate quantile noise instead of hand-tuned Student-t noise | OT flow matching · HRRR-mini precipitation | tail KS distance: 0.1014 -> 0.0393 | Improved |
| replace Gaussian noise with heavy-tailed noise (Student-t) (ablation) | flow matching · HRRR-mini precipitation | kurtosis deviation: 4.930 -> 2.890 | Improved |
| learn quantile noise instead of Gaussian on a larger model | OT flow matching, 55M U-Net · CIFAR-10 | FID: 3.37 -> 3.25 | Improved |
| learn quantile noise instead of Gaussian on 35M U-Net | OT flow matching, 35M U-Net · CIFAR-10 | FID | No clear effect |
| learned quantile noise at reduced U-Net capacity | OT flow matching · MNIST (channels 8/16/32) | FID | Improved |
| use learned data-adapted latent instead of Gaussian with OT coupling | OT flow matching · checkerboard | transport path length | Improved |
| use learned latent for heavy-tailed funnel instead of Gaussian or compact noise | flow matching, small network · Neal's funnel | sample quality (qualitative) | Improved |
| add stop-gradient on target displacement for the quantile (ablation) | quantile network training · joint training | collapse avoidance | Improved |
Closest papers by meaning
- Annealing Flow Generative Models Towards Sampling High-Dimensional and Multi-Modal Distributions (ICML 2025 · similarity 0.87)
- Analysis of Learning a Flow-based Generative Model from Limited Sample Complexity (ICLR 2024 · similarity 0.86)
- Towards Non-Asymptotic Convergence for Diffusion-Based Generative Models (ICLR 2024 · similarity 0.86)
- Information Theoretic Learning for Diffusion Models with Warm Start (NeurIPS 2025 · similarity 0.85)
- Diffusion Models With Learned Adaptive Noise (NeurIPS 2024 · similarity 0.85)
- Latent Bayesian Optimization via Autoregressive Normalizing Flows (ICLR 2025 · similarity 0.85)
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.
