EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

A statistical physics framework for optimal learning

Francesca Mignacco iD, Francesco Mori

PNAS Nexus · 30.05.2026 · 5 (6)

DOI: 10.1093/pnasnexus/pgag182 ↗

Kurz gefasst

Fachartikel in der Zeitschrift PNAS Nexus, online erschienen am 26.05.2026. Verfasst von Francesca Mignacco und 1 weiteren Autor:innen. Beteiligte Einrichtung: Joseph Henry Laboratories of Physics, Princeton University, Princeton, NJ 08544, USA. Die Arbeit nennt 77 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Simons Foundation, Human Frontier Science Program, Leverhulme Trust.

AbstractSchlagwörterBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Learning is a complex dynamical process shaped by a range of interconnected decisions. Careful design of hyperparameter schedules for artificial neural networks or efficient allocation of cognitive resources by biological learners can dramatically affect performance. Yet, theoretical understanding of optimal learning strategies remains sparse, especially due to the intricate interplay between evolving metaparameters and nonlinear learning dynamics. The search for optimal protocols is further hindered by the high dimensionality of the learning space, often resulting in predominantly heuristic, difficult to interpret, and computationally demanding solutions. Here, we combine statistical physics with control theory in a unified theoretical framework to identify optimal learning protocols in prototypical neural network models. In the high-dimensional limit, we derive closed-form ordinary differential equations that track online stochastic gradient descent through low-dimensional order parameters. We formulate the design of learning protocols as an optimal control problem directly on the dynamics of the order parameters with the goal of minimizing the generalization error. This formulation encompasses a variety of learning scenarios, optimization constraints, and control budgets. We apply it to representative cases, including optimal curricula, adaptive dropout regularization and noise schedules in denoising autoencoders. We find nontrivial yet interpretable strategies highlighting how optimal protocols mediate learning trade-offs. Our results establish a principled foundation for understanding and designing optimal protocols and suggest a path toward a theory of meta-learning grounded in statistical physics.

Schlagwörter

Stichwörter der Autor:innen

meta-learningoptimal controlstatistical physicstraining protocols

Autor:innen und Institutionen

  • Joseph Henry Laboratories of Physics, Princeton University, Princeton, NJ 08544, USA.
  • Graduate Center, City University of New York, New York, NY 10016, USA.
  • Center of Mathematical Sciences and Applications, Harvard University, Cambridge, MA 02138, USA.

ORCID-Profile: Francesca Mignacco

Förderung

  • Simons Foundation
  • Human Frontier Science Program
  • Leverhulme Trust

Zitieren

APA

Mignacco, F., & Mori, F. (2026). A statistical physics framework for optimal learning. PNAS Nexus, 5(6). https://doi.org/10.1093/pnasnexus/pgag182

Vancouver

Mignacco F, Mori F. A statistical physics framework for optimal learning. PNAS Nexus. 2026;5(6). doi:10.1093/pnasnexus/pgag182

BibTeX
@article{Mignacco2026,
  author = {Mignacco, Francesca and Mori, Francesco},
  title = {A statistical physics framework for optimal learning},
  journal = {PNAS Nexus},
  year = {2026},
  volume = {5},
  number = {6},
  doi = {10.1093/pnasnexus/pgag182}
}
Kostenlos starten →