Neural networks - on awarding the Nobel Prize in Physics 2024

O. S. Savchuk^a, I. V. Orlov<sup>b

a</sup> Moscow Institute of Physics and Technology (National Research University), Dolgoprudny, Moscow Region
^b V. I. Vernadsky Crimean Federal University, Simferopol

Abstract: This article provides a comprehensive overview of the 2024 Nobel Prize in Physics, awarded to John Hopfield and Geoffrey Hinton for their pioneering work in integrating fundamental physics principles into machine learning and neural networks. The paper explores the profound connection between physics and machine learning, emphasizing the key physical concepts that underpin the neural network models developed by these laureates. The article begins with brief biographies of Hopfield and Hinton, highlighting their significant scientific contributions. John Hopfield is renowned for his development of the Hopfield network, a type of recurrent artificial neural network, while Geoffrey Hinton is celebrated as the «godfather of AI» for his work on artificial neural networks, including the backpropagation method and the Boltzmann machine. The paper then delves into the relationship between physics and machine learning, focusing on statistical physics, energy landscapes, and chaos theory. Statistical physics offers a theoretical framework for understanding complex systems, while energy landscapes assist algorithms in finding optimal solutions. Chaos theory elucidates the sensitivity of systems to initial conditions, crucial for models like recurrent neural networks. Hopfield's associative memory model, inspired by atomic spin concepts, enables the storage and retrieval of patterns, akin to the human brain's associative memory. Hinton's model builds on Hopfield's ideas, incorporating the Boltzmann machine concept and focusing on the probability of neuron activation states. Despite the Nobel Committee's recognition, the article acknowledges the historical roots of neural networks, tracing back to early models such as McCulloch and Pitts' threshold logic and Hebb's neuroplasticity concept. The paper concludes by assessing the real impact of Hopfield and Hinton's contributions, noting that while they did not invent neural networks, their work revitalized the field, paving the way for modern architectures like deep convolutional networks. Their research continues to influence the integration of artificial intelligence with other scientific domains, including physics, suggesting potential applications in predicting molecular properties, developing solar cells, and measuring gravitational waves. This article underscores the enduring significance of their contributions to the advancement of artificial intelligence and neural networks.

Keywords: neural network, associative memory, Hopfield neural network, Hinton neural network, Boltzmann machine, spin glasses, energy landscapes, pattern

UDC: 004.8, 51-73, 53.07

MSC: 68T07, 68U07, 68U10, 68U15