MetaDNS: Enhancing Exploration in Discrete Neural Samplers via Well-Tempered Metadynamics
Xiaochen Du, Juno Nam, Jaemoo Choi, Wei Guo, Sathya Edamadaka, Junyi Sha, Elton Pan, Yongxin Chen, Molei Tao, Rafael Gómez-Bombarelli · 2026
Read MetaDNS: Enhancing Exploration in Discrete Neural Samplers via Well-Tempered Metadynamics by Xiaochen Du, Juno Nam, Jaemoo Choi, Wei Guo, Sathya Edamadaka, Junyi Sha, Elton Pan, Yongxin Chen, Molei Tao, Rafael Gómez-Bombarelli (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-2605-21722.txt.
Description
Du and colleagues address sampling from discrete distributions with many modes split by high energy barriers. Neural samplers collapse onto single modes and miss barrier regions, which matters for free energy figures and phase changes. Their MetaDNS framework weaves well-tempered metadynamics into discrete diffusion or autoregressive samplers. An adaptive history-dependent bias pushes search into once closed-off ground. On Ising, Potts, and copper-gold alloy benchmarks it matches the correct thermodynamic distribution and free energy values, with fewer bias steps than Markov-chain metadynamics.
Catalog record
| CREATOR | Xiaochen Du, Juno Nam, Jaemoo Choi, Wei Guo, Sathya Edamadaka, Junyi Sha, Elton Pan, Yongxin Chen, Molei Tao, Rafael Gómez-Bombarelli |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 79225 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2605.21722