Research map / Probabilistic and scientific modeling / Variational and generative sampling
Smooth Normalizing Flows
Jonas Köhler, Andreas Krämer, Frank Noe
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 smooth mixture transformations instead of neural spline transformations | Alanine dipeptide flow with pure density estimation · Alanine dipeptide test set | FME × 10^4: 12.13 -> 1.04 | Improved |
| Use smooth mixture transformations instead of neural spline transformations | Alanine dipeptide flow with pure density estimation · Alanine dipeptide test set | NLL: -210.32 -> -211.04 | No clear effect |
| Add force matching loss (ablation) | Smooth flow · Alanine dipeptide test set; ωk=0 | FME × 10^4: 1.04 -> 0.32 | Improved |
| Add force matching loss (ablation) | Smooth flow · Alanine dipeptide test set; ωk=0 | NLL: -211.04 -> -211.40 | Improved |
| Add reverse KLD loss (ablation) | Smooth flow · Alanine dipeptide test set | NLL: -211.04 -> -206.33 | Hurt |
| Add reverse KLD loss (ablation) | Smooth flow · Alanine dipeptide test set | KLD: 193.91 -> 219.03 | Mixed |
| Add reverse KLD and force matching losses (ablation) | Smooth flow · Alanine dipeptide test set | FME × 10^4: 1.04 -> 0.48 | Mixed |
| Train the spline flow with force matching loss (ablation) | Spline flow · Alanine dipeptide training runs; ωk=0 | successful runs: 10/10 failed runs | Hurt |
| Train the spline flow with force matching loss (ablation) | Spline flow · Alanine dipeptide training runs; ωk=0.1 | successful runs: 7/10 failed runs | Hurt |
| Use smooth transformations as the flow potential rather than spline transformations | Alanine dipeptide molecular dynamics · Microcanonical MD | energy stability: 8×10^{-4} kJ/mol vs 6×10^{-4} kJ/mol | Improved |
| Replace analytic inverse evaluation with multi-bin root finding | Neural spline flow, dim 2 · Runtime benchmark | inversion runtime factor: 2.1 | Hurt |
| Replace analytic inverse evaluation with multi-bin root finding | Neural spline flow, dim 2048 · Runtime benchmark | inversion runtime factor: 13.4 | Hurt |
Closest papers by meaning
- Conditional Normalizing Flows for Active Learning of Coarse-Grained Molecular Representations (ICML 2024 · similarity 0.90)
- Path Gradients after Flow Matching (NeurIPS 2025 · similarity 0.89)
- Scalable Equilibrium Sampling with Sequential Boltzmann Generators (ICML 2025 · similarity 0.88)
- AlphaFold Meets Flow Matching for Generating Protein Ensembles (ICML 2024 · similarity 0.86)
- Energy-Based Flow Matching for Generating 3D Molecular Structure (ICML 2025 · similarity 0.86)
- Temperature-Annealed Boltzmann Generators (ICML 2025 · 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.
