Research map / Probabilistic and scientific modeling / Variational and generative sampling
Normalizing flow neural networks by JKO scheme
Chen Xu, Xiuyuan Cheng, Yao Xie
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 |
|---|---|---|---|
| train with JKO-iFlow instead of OT-Flow | normalizing flow models · POWER | NLL: 0.32 -> -0.12 | Improved |
| train with JKO-iFlow instead of OT-Flow | normalizing flow models · GAS | NLL: -6.04 -> -7.65 | Improved |
| train with JKO-iFlow instead of OT-Flow | normalizing flow models · MINIBOONE | NLL: 11.44 -> 12.55 | Hurt |
| train with JKO-iFlow instead of OT-Flow | normalizing flow models · BSDS300 | NLL: -104.62 -> -153.82 | Improved |
| use block-wise JKO-iFlow training instead of OT-Flow | POWER · POWER | training time (h): 0.36 -> 0.07 | Improved |
| use block-wise JKO-iFlow training instead of OT-Flow | GAS · GAS | training time (h): 0.23 -> 0.07 | Improved |
| use block-wise JKO-iFlow training instead of OT-Flow | MINIBOONE · MINIBOONE | training time (h): 0.21 -> 0.03 | Improved |
| use block-wise JKO-iFlow training instead of OT-Flow | BSDS300 · BSDS300 | training time (h): 0.62 -> 0.05 | Improved |
| add trajectory reparameterization (ablation) | JKO-iFlow · MINIBOONE | NLL: 12.55 to 10.55 | Improved |
| use 3 residual blocks instead of 4 (ablation) | JKO-iFlow · POWER | NLL: -0.12 -> -0.06 | Hurt |
| use 5 residual blocks instead of 4 (ablation) | JKO-iFlow · GAS | NLL: -7.65 -> -7.80 | Improved |
| increase batch size from 1000 to 5000 | JKO-iFlow · BSDS300 | NLL: -153.82 -> -156.68 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- OT-Flow — OT-Flow: Fast and Accurate Continuous Normalizing Flows via Optimal Transport (AAAI 2021) · better on 7 of 8 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Importance Corrected Neural JKO Sampling (ICML 2025 · similarity 0.88)
- Scalable Wasserstein Gradient Flow for Generative Modeling through Unbalanced Optimal Transport (ICML 2024 · similarity 0.86)
- An Error Analysis of Flow Matching for Deep Generative Modeling (ICML 2025 · similarity 0.86)
- Annealing Flow Generative Models Towards Sampling High-Dimensional and Multi-Modal Distributions (ICML 2025 · similarity 0.85)
- Markovian Flow Matching: Accelerating MCMC with Continuous Normalizing Flows (NeurIPS 2024 · similarity 0.84)
- Fast and unified path gradient estimators for normalizing flows (ICLR 2024 · similarity 0.84)
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.
