Dynamical mean-field analysis of adaptive Langevin diffusions: Replica-symmetric fixed point and empirical Bayes
Zhou Fan, Justin Ko, Bruno Loureiro, Yue M. Lu, Yandi Shen · 2025
Read Dynamical mean-field analysis of adaptive Langevin diffusions: Replica-symmetric fixed point and empirical Bayes by Zhou Fan, Justin Ko, Bruno Loureiro, Yue M. Lu, Yandi Shen (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-2504-15558.txt.
Introduction
Some samplers change as they run, because the prior they draw from keeps learning from the samples so far. Fan, Ko, Loureiro, Lu, and Shen study one such scheme: a Langevin diffusion that samples a Bayesian linear model while its prior keeps updating by a fit to the samples seen. A companion paper set out the high-dimensional limit for how the prior and the samples move together.
Here the authors study that limit when the posterior is smooth enough to mix well. They show the sampler settles, on a time scale that does not grow with dimension, to a fixed state given by a small set of scalar equations. The prior lands on a key point of a simpler free-energy formula, and the work gives a fresh proof that the simpler formula is exact.
Catalog record
| CREATOR | Zhou Fan, Justin Ko, Bruno Loureiro, Yue M. Lu, Yandi Shen |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 335 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2504-15558