No free lunch for stochastic gradient Langevin dynamics
Journal of Applied Probability · 15.06.2026 · 1-17
DOI: 10.1017/jpr.2026.10109 ↗Kurz gefasst
Fachartikel in der Zeitschrift Journal of Applied Probability, online erschienen am 15.06.2026. Verfasst von Natesh Pillai und 2 weiteren Autor:innen. Die Arbeit nennt 21 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
As sample sizes grow, scalability has become a central concern in the development of Markov chain Monte Carlo (MCMC) methods. One general approach to this problem, exemplified by the popular stochastic gradient Langevin dynamics (SGLD) algorithm, is to use a small random subsample of the data at every time step. This paper shows that this approach often fails: while decreasing the sample size increases the speed of each MCMC step, for typical datasets this is balanced by a matching decrease in accuracy. This result complements recent work on lower bounds for subsampling MCMC by focusing on actual errors for SGLD, including non-reversible algorithms, and by giving assumptions that are comparatively easy to verify.
Zitieren
APA
Pillai, N., Smith, A., & Zaman, A. (2026). No free lunch for stochastic gradient Langevin dynamics. Journal of Applied Probability, 1-17. https://doi.org/10.1017/jpr.2026.10109
Vancouver
Pillai N, Smith A, Zaman A. No free lunch for stochastic gradient Langevin dynamics. Journal of Applied Probability. 2026:1-17. doi:10.1017/jpr.2026.10109
BibTeX
@article{Pillai2026,
author = {Pillai, Natesh and Smith, Aaron and Zaman, Azeem},
title = {No free lunch for stochastic gradient Langevin dynamics},
journal = {Journal of Applied Probability},
year = {2026},
pages = {1--17},
doi = {10.1017/jpr.2026.10109}
}
EUVIMED