Research map / Probabilistic and scientific modeling / Variational and generative sampling
Efficient Regression-based Training of Normalizing Flows for Boltzmann Generators
Danyal Rehman, Oscar Davis, Jiarui Lu, Jian Tang, Michael M. Bronstein, Yoshua Bengio, Alexander Tong, Joey Bose
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 RegFlow instead of MLE | NSF · Alanine dipeptide | E-W1: 13.797 -> 0.519 | Improved |
| train with RegFlow instead of MLE | NSF · Alanine tetrapeptide | E-W1: 20.886 -> 6.277 | Improved |
| train with RegFlow instead of MLE | NSF · Alanine tripeptide | T-W2: 1.665 -> 1.612 | Improved |
| train with RegFlow instead of MLE | NSF · Alanine dipeptide | ESS: 0.055 -> 0.036 | Hurt |
| train with RegFlow instead of MLE | Res-NVP · Alanine dipeptide | E-W1: >10^3 -> 2.310 | Improved |
| train with RegFlow instead of MLE | Jet · Alanine dipeptide | E-W1: >10^3 -> 6.349 | Improved |
| train with RegFlow instead of MLE | Jet · Alanine tetrapeptide | E-W1: >10^3 -> >10^3 | No clear effect |
| train with RegFlow instead of MLE (likelihood inference) | NSF · Alanine dipeptide | inference time: 277.00 -> 8.18 | Improved |
| train with RegFlow instead of MLE (likelihood inference) | Res-NVP · Alanine dipeptide | inference time: 3.64 -> 3.51 | No clear effect |
| use RegFlow instead of MLE to reach same E-W1 | NSF · Alanine dipeptide | training time: 10h10 -> 6h54 | Improved |
| add regularization (logdet) to RegFlow (ablation) | NSF · Alanine dipeptide | E-W1: 0.604 -> 0.519 | Improved |
| add regularization (logdet) to RegFlow (ablation) | Jet · Alanine dipeptide | E-W1: 9.707 -> 6.349 | Improved |
Closest papers by meaning
- Path Gradients after Flow Matching (NeurIPS 2025 · similarity 0.89)
- Scalable Equilibrium Sampling with Sequential Boltzmann Generators (ICML 2025 · similarity 0.89)
- BoltzNCE: Learning likelihoods for Boltzmann Generation with Stochastic Interpolants and Noise Contrastive Estimation (NeurIPS 2025 · similarity 0.89)
- Efficient Molecular Conformer Generation with SO(3)-Averaged Flow Matching and Reflow (ICML 2025 · similarity 0.88)
- Conditional Normalizing Flows for Active Learning of Coarse-Grained Molecular Representations (ICML 2024 · similarity 0.88)
- AlphaFold Meets Flow Matching for Generating Protein Ensembles (ICML 2024 · similarity 0.88)
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.
