BoltzNCE: Learning Likelihoods for Boltzmann Generation with Stochastic Interpolants and Noise Contrastive Estimation
Rishal Aggarwal, Jacky Chen, Nicholas M. Boffi, David Ryan Koes · 2025
Read BoltzNCE: Learning Likelihoods for Boltzmann Generation with Stochastic Interpolants and Noise Contrastive Estimation by Rishal Aggarwal, Jacky Chen, Nicholas M. Boffi, David Ryan Koes (2025) 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-2507-00846.txt.
Description
Rishal Aggarwal and co-authors make a costly step in molecule simulation a hundred times cheaper. Their 2025 paper concerns Boltzmann generators, which learn to sample molecular shapes with the right odds. Correct odds require each sample's likelihood, but computing it normally means tracking expensive Jacobian terms during integration.
Their system, BoltzNCE, trains an energy-based model to estimate those likelihoods instead. It blends two training goals, noise contrastive estimation and score matching, which together beat either alone. On a small test molecule, the estimated odds reproduce exact free energy profiles closely at a hundredth of the cost. Training on several small peptides even transferred to new ones, running at least six times faster than standard molecular dynamics.
Catalog record
| CREATOR | Rishal Aggarwal, Jacky Chen, Nicholas M. Boffi, David Ryan Koes |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 43 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2507-00846