Few-Step Boltzmann Generators via Scalable Likelihood Flow Maps
RuiKang OuYang ∗, Hanlin Yu †, Xinyue Ai †, Yutong He, Nicholas M. Boffi, Pradeep Ravikumar, José Miguel Hernández-Lobato, Max Simchowitz, Benjamin Kurt Miller, Omar Chehab · 2026
Read Few-Step Boltzmann Generators via Scalable Likelihood Flow Maps by RuiKang OuYang ∗, Hanlin Yu †, Xinyue Ai †, Yutong He, Nicholas M. Boffi, Pradeep Ravikumar, José Miguel Hernández-Lobato, Max Simchowitz, Benjamin Kurt Miller, Omar Chehab (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2606-29110.txt.
Description
OuYang and colleagues introduce SCALLOP, a method for training Boltzmann generators, models that sample molecular configurations with the right physical probabilities. Flow-based generators can produce high-quality samples in few steps, but estimating the model's likelihood, how probable each sample is, has lagged. Existing approaches need restrictive architectures for exact calculation, or noisy trace estimators with high variance.
SCALLOP builds on a likelihood flow map model that generates samples and their densities in few steps. It replaces that model's noisy training estimator with a new distillation objective that avoids it entirely and admits a vectorised form. In tests, SCALLOP cut training variance and time while beating its predecessor. It stayed competitive with the state of the art on images too. Inference ran up to ten times faster than the fastest baseline.
Catalog record
| CREATOR | RuiKang OuYang ∗, Hanlin Yu †, Xinyue Ai †, Yutong He, Nicholas M. Boffi, Pradeep Ravikumar, José Miguel Hernández-Lobato, Max Simchowitz, Benjamin Kurt Miller, Omar Chehab |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 85 KB |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2606.29110