Saved in:
| Main Authors: | , , , , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2026
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2605.21722 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866913151128174592 |
|---|---|
| author | Du, Xiaochen Nam, Juno Choi, Jaemoo Guo, Wei Edamadaka, Sathya Sha, Junyi Pan, Elton Chen, Yongxin Tao, Molei Gómez-Bombarelli, Rafael |
| author_facet | Du, Xiaochen Nam, Juno Choi, Jaemoo Guo, Wei Edamadaka, Sathya Sha, Junyi Pan, Elton Chen, Yongxin Tao, Molei Gómez-Bombarelli, Rafael |
| contents | Sampling from discrete distributions with multiple modes and energy barriers is fundamental to machine learning and computational physics. Recent discrete neural samplers like MDNS suffer from mode collapse and fail to sample high-energy barrier regions between modes, which is critical for free energy estimation and understanding phase transitions. We propose Metadynamics Discrete Neural Sampler (MetaDNS), a general framework integrating well-tempered metadynamics into discrete diffusion or autoregressive samplers. By maintaining an adaptive, history-dependent bias potential along selected low-dimensional coordinates, MetaDNS forces exploration of previously inaccessible regions, enabling free energy reconstruction infeasible with standard neural samplers due to a lack of high-energy samples. On challenging low-temperature benchmarks including Ising, Potts, and the copper-gold binary alloy, MetaDNS reproduces the thermodynamic distribution. Compared to MCMC-based metadynamics, MetaDNS also achieves comparable exploration requiring fewer bias deposition steps. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2605_21722 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | MetaDNS: Enhancing Exploration in Discrete Neural Samplers via Well-Tempered Metadynamics Du, Xiaochen Nam, Juno Choi, Jaemoo Guo, Wei Edamadaka, Sathya Sha, Junyi Pan, Elton Chen, Yongxin Tao, Molei Gómez-Bombarelli, Rafael Statistical Mechanics Materials Science Machine Learning I.2.6; J.2 Sampling from discrete distributions with multiple modes and energy barriers is fundamental to machine learning and computational physics. Recent discrete neural samplers like MDNS suffer from mode collapse and fail to sample high-energy barrier regions between modes, which is critical for free energy estimation and understanding phase transitions. We propose Metadynamics Discrete Neural Sampler (MetaDNS), a general framework integrating well-tempered metadynamics into discrete diffusion or autoregressive samplers. By maintaining an adaptive, history-dependent bias potential along selected low-dimensional coordinates, MetaDNS forces exploration of previously inaccessible regions, enabling free energy reconstruction infeasible with standard neural samplers due to a lack of high-energy samples. On challenging low-temperature benchmarks including Ising, Potts, and the copper-gold binary alloy, MetaDNS reproduces the thermodynamic distribution. Compared to MCMC-based metadynamics, MetaDNS also achieves comparable exploration requiring fewer bias deposition steps. |
| title | MetaDNS: Enhancing Exploration in Discrete Neural Samplers via Well-Tempered Metadynamics |
| topic | Statistical Mechanics Materials Science Machine Learning I.2.6; J.2 |
| url | https://arxiv.org/abs/2605.21722 |