Techniques from physics

The win is already causing a stir on social media, as it seems unusual that research in a computer science field like machine learning could win a Nobel Prize in Physics. “And the 2024 Nobel Prize in Physics does not go to physics…” German physicist Sabine Hossenfelder tweeted this morning.

From the Nobel Committee’s perspective, the award is based largely on the fact that the two men drew on statistical models of physics and partly on recognition of the advances in physics research that resulted from the use of the men’s neural network techniques as research tools.

Nobel Committee Chairwoman Ellen Moons, a physicist at Karlstad University in Sweden, said during the announcement: “Artificial neural networks have been used to advance research in physics topics as diverse as particle physics, materials science and astrophysics.”

Hopfield, a 91-year-old theoretical biologist with a background in physics, made a breakthrough in 1982 by developing a network that described connections between nodes as physical forces, as Nature describes it in a report. His innovation, known as the Hopfield network, uses concepts from physics that describe how atomic spins behave in materials. Specifically, it stores patterns as low-energy states, allowing the system to recreate images when prompted with similar patterns. This approach mimicked associative memory and was similar to the way the brain recalls words or concepts.

Hinton, 76, built on Hopfield’s research in the early 1980s by developing a multilayer version of the Hopfield network that incorporated probabilities. Hinton drew parallels with physical studies of large systems of similar elements such as gas molecules. Instead of tracking individual molecules, physicists study collective properties such as pressure or temperature. The Boltzmann equation from 19th century physics calculates the probability of different states in such systems. Hinton applied this concept to neural networks and called his 1985 method the “Boltzmann machine,” which highlighted the connection between machine learning and statistical physics. A Boltzmann machine is able to recognize and classify images and generate new examples based on its training data.