Thermodynamic learning
Abstract
We discuss the possibility to train a thermodynamic system, whose micro-variables are fully determined through the Hamiltonian by the usual statistical mechanical rules, to perform tasks such as memorization and generalization. Training is achieved by the application of suitable external fields, playing the role of {\it data}. At variance with conventional machine learning, no other logical or algorithmic rules are introduced. The system is amenable, in principle, to exact analytical calculations. We specialize this general approach to a prototypical Ising system with annealed dichotomous couplings and study its learning ability. Results indicate excellent memorization and good generalization capacity already for small systems, and a tendency to improve with system size.
Related
- Online local learning for generative thermodynamic computing
- Hamiltonian learning reveals optoelectronic mechanisms across thermodynamic state space in soft semiconductors
- Charting the thermodynamic stability of hybrid perovskite alloys with machine learning
- A thermodynamic correction layer for machine learning interatomic potentials
- Machine-Learning Exploration of Defect Topologies and Thermodynamic Stability in Graphene with Atomic Vacancies
BibTeX
@misc{corberi2026thermodynamic,
title = {{Thermodynamic learning}},
author = {F. Corberi and S. dello Russo and G. Messuti and S. Scarpetta and L. Smaldone},
howpublished = {arXiv},
year = {2026},
month = sep,
doi = {10.48550/arxiv.2609.04732},
eprint = {2609.04732},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2609.04732},
}
RIS
TY - UNPB
TI - Thermodynamic learning
AU - F. Corberi
AU - S. dello Russo
AU - G. Messuti
AU - S. Scarpetta
AU - L. Smaldone
PY - 2026
DA - 2026/09/10
PB - arXiv
DO - 10.48550/arxiv.2609.04732
UR - https://arxiv.org/abs/2609.04732
AB - We discuss the possibility to train a thermodynamic system, whose micro-variables are fully determined through the Hamiltonian by the usual statistical mechanical rules, to perform tasks such as memorization and generalization. Training is achieved by the application of suitable external fields, playing the role of {\it data}. At variance with conventional machine learning, no other logical or algorithmic rules are introduced. The system is amenable, in principle, to exact analytical calculations. We specialize this general approach to a prototypical Ising system with annealed dichotomous couplings and study its learning ability. Results indicate excellent memorization and good generalization capacity already for small systems, and a tendency to improve with system size.
ER -