Research map / Probabilistic and scientific modeling / Variational and generative sampling
Diffusion Normalizing Flow
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| reduce sampling steps N from 100 to 5 (ablation) | DiffFlow · CIFAR-10 | FID: 13.43 -> 28.31 | Hurt |
| reduce sampling steps N from 100 to 5 | DDPM (L_s) · CIFAR-10 | FID: 10.04 -> 370.23 | Hurt |
| reduce sampling steps N from 100 to 5 | DDIM · CIFAR-10 | FID: 5.63 -> 44.69 | Hurt |
| relative FID degeneracy at N=5 relative to N=100 | DiffFlow vs DDPM (L) · CIFAR-10 | relative FID ratio: 2.12 vs 16.80 | Helped |
| use flexible random timestamps L_beta hat instead of fixed L_beta (ablation) | DiffFlow · CIFAR-10 | FID: 13.87 -> 13.43 | Helped |
| use flexible random timestamps L_beta hat instead of fixed L_beta (ablation) | DiffFlow · CIFAR-10 | NLL: 3.71 -> 3.67 | Helped |
| evaluate NLL with probability-flow ODE instead of SDE ELBO (ablation) | DiffFlow (L_beta hat) · CIFAR-10 | NLL: 3.67 -> 3.04 | Helped |
| evaluate with probability-flow ODE instead of SDE (ablation) | DiffFlow (L_beta hat) · CIFAR-10 | FID: 13.43 -> 14.14 | Hurt |
| DiffFlow (ODE) vs FFJORD, noise added in training | DiffFlow · POWER | NLL (nats): -0.46 -> -1.04 | Helped |
| DiffFlow (ODE) vs FFJORD | DiffFlow · HEPMASS | NLL (nats): 14.92 -> 15.04 | No clear effect |
| DiffFlow (ODE) vs NAF | DiffFlow · GAS | NLL (nats): -11.96 -> -10.45 | Hurt |
| DiffFlow (ODE) vs NAF | DiffFlow · MINIBOONE | NLL (nats): 8.86 -> 8.06 | Helped |
Closest papers by meaning
- Normalizing Flows are Capable Generative Models (ICML 2025 · similarity 0.87)
- Neural Flow Diffusion Models: Learnable Forward Process for Improved Diffusion Modelling (NeurIPS 2024 · similarity 0.87)
- On the Trajectory Regularity of ODE-based Diffusion Sampling (ICML 2024 · similarity 0.87)
- Neural Stochastic Flows: Solver-Free Modelling and Inference for SDE Solutions (NeurIPS 2025 · similarity 0.86)
- Neural Diffusion Models (ICML 2024 · similarity 0.86)
- Denoising Diffusion Bridge Models (ICLR 2024 · similarity 0.86)
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.