Efficient Training of Boltzmann Generators Using Off-Policy Log-Dispersion Regularization
Henrik Schopmans, Christopher von Klitzing, Pascal Friederich · 2026
Read Efficient Training of Boltzmann Generators Using Off-Policy Log-Dispersion Regularization by Henrik Schopmans, Christopher von Klitzing, Pascal Friederich (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-2602-03729.txt.
Description
Henrik Schopmans and 2 co-authors improve how Boltzmann generators learn, in a 2026 paper on sampling from physical energy landscapes. These generative models draw independent samples from a system's Boltzmann distribution, but training them is data-hungry: simulations and energy evaluations are both expensive.
Their off-policy log-dispersion regularization builds on a generalized log-variance objective and works alongside standard data-based training. It uses target energy labels as extra information to shape the learned energy landscape, without demanding fresh on-policy samples. The framework accepts biased or unbiased simulation data, and even purely variational training with no target samples at all. Across benchmarks it improves both data efficiency and final performance, with sample-efficiency gains up to tenfold.
Catalog record
| CREATOR | Henrik Schopmans, Christopher von Klitzing, Pascal Friederich |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 99 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-03729