Non-Equilibrium Model Selection via Finite-Time Thermodynamics
Journal of the Physical Society of Japan · 15.08.2026 · 95 (8)
DOI: 10.7566/jpsj.95.083002 ↗Kurz gefasst
Fachartikel in der Zeitschrift Journal of the Physical Society of Japan, erschienen am 15.08.2026. Verfasst von Masayuki Ohzeki. Die Arbeit nennt 12 Literaturangaben. Gefördert von: Cabinet Office, Government of Japan.
Abstract (Originalsprache)
Information criteria such as WAIC and WBIC extend model selection to singular learning machines, but they are usually derived for equilibrium posteriors. We formulate a finite-time analogue of WBIC by replacing the equilibrium posterior with an effective ensemble generated by learning dynamics under a resource constraint. When this ensemble admits an analytic effective potential, singular learning theory yields a resource-dependent real log canonical threshold. The resulting estimator gives a computable thermodynamic contribution to time-bounded MDL and identifies the finite-time singular complexity relevant to the structural information measured by epiplexity.
Förderung
- Cabinet Office, Government of Japan
Zitieren
APA
Ohzeki, M. (2026). Non-Equilibrium Model Selection via Finite-Time Thermodynamics. Journal of the Physical Society of Japan, 95(8). https://doi.org/10.7566/jpsj.95.083002
Vancouver
Ohzeki M. Non-Equilibrium Model Selection via Finite-Time Thermodynamics. Journal of the Physical Society of Japan. 2026;95(8). doi:10.7566/jpsj.95.083002
BibTeX
@article{Ohzeki2026,
author = {Ohzeki, Masayuki},
title = {Non-Equilibrium Model Selection via Finite-Time Thermodynamics},
journal = {Journal of the Physical Society of Japan},
year = {2026},
volume = {95},
number = {8},
doi = {10.7566/jpsj.95.083002}
}
EUVIMED