Physical Society Colloquium

Is Machine Learning Physics?

Paul François

Département de biochimie et médecine moléculaire
Université de Montréal

The 2024 Physics Nobel Prize was awarded to John J. Hopfield and Geoffrey Hinton "for foundational discoveries and inventions that enable machine learning with artificial neural networks’’. This talk will provide an accessible overview of their key innovations recognized by the Nobel Prize—Hopfield Networks and Restricted Boltzmann Machines—within the framework of statistical mechanics.  Specifically, I will show how Hopfield Networks leverage concepts from energy landscapes and spin glasses. I will also show how Restricted Boltzmann Machines can be derived and generalized from Hopfield Networks, including a derivation of so-called hidden fields. Finally, we will connect these foundational ideas to modern machine learning architectures, from multilayered neural networks to transformers, which are the backbone of systems like ChatGPT.