Research map / Probabilistic and scientific modeling / Variational and generative sampling
Learning Escorted Protocols For Multistate Free-Energy Estimation
Lars Holdijk, Nithishwer Mouroug Anand, Michael M. Bronstein, Max Welling
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 escorted protocol flow graph (EPFG) concatenation instead of pairwise summation (ablation) | E-NEQ with underdamped Langevin · ADP | MAE: 0.93 -> 0.88 | Improved |
| use escorted protocol flow graph (EPFG) concatenation instead of pairwise summation (ablation) | E-NEQ with overdamped Langevin · ADP | MAE: 0.96 -> 0.59 | Improved |
| use escorted protocol flow graph (EPFG) concatenation instead of pairwise summation (ablation) | E-NEQ with HMC · ADP | MAE: 0.99 -> 0.95 | Improved |
| learn escorting field and potential (E-NEQ UL) instead of TFEP vector field only | TFEP · ADP | MAE: 1.17 -> 0.93 | Improved |
| learn escorting field and potential (E-NEQ OL with EPFG) instead of TFEP | TFEP · ADP | MAE: 1.17 -> 0.59 | Improved |
| use E-NEQ instead of TFEP for free-energy correlation with US | TFEP · ADP | R2 / Pearson r: R2 0.90 -> 0.91, r 0.95 -> 0.96 | Improved |
| add the escort vector field to JE with learned potential (E-NEQ 100 steps vs JE 50,000 steps) (ablation) | JE with learned potential · ADP | MAE: 3.92 -> 0.93 | Improved |
| increase JE interpolation steps from 500 to 5,000 | JE with learned potential · ADP | MAE: 6.16 -> 7.73 | Hurt |
| increase JE interpolation steps from 5,000 to 50,000 | JE with learned potential · ADP | MAE: 7.73 -> 3.92 | Improved |
| use Langevin dynamics rather than HMC as stationary-preserving component | E-NEQ · ADP | successful trajectories from C5: 32477 -> 48328 | Improved |
| use overdamped Langevin instead of HMC for stationary-preserving dynamics | E-NEQ · ADP | MAE with EPFG: 0.95 -> 0.59 | Improved |
| add Metropolis-Hastings correction against HMC divergence | E-NEQ HMC · ADP | trajectory stability | No clear effect |
Closest papers by meaning
- FEAT: Free energy Estimators with Adaptive Transport (NeurIPS 2025 · similarity 0.87)
- Free energy Estimation on Any State Space (NeurIPS 2026 · similarity 0.85)
- Pairwise Optimal Transports for Training All-to-All Flow-Based Condition Transfer Model (NeurIPS 2025 · similarity 0.84)
- Iterated Denoising Energy Matching for Sampling from Boltzmann Densities (ICML 2024 · similarity 0.84)
- On scalable and efficient training of diffusion samplers (NeurIPS 2025 · similarity 0.84)
- Action-Minimization Meets Generative Modeling: Efficient Transition Path Sampling with the Onsager-Machlup Functional (ICML 2025 · 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.
