The 2024 Nobel Prize in Physics was announced on October 8 at around 5:45 p.m. BST. John J. Hopfield, a scientist at Princeton University in the United States, and Geoffrey E. Hinton, a scientist at the University of Toronto in Canada, were awarded the prize for their "fundamental discoveries and inventions based on artificial neural networks that enable machine learning."

They use physics to train artificial neural networks

Two of this year's Nobel Laureates in Physics used the tools of physics to develop methods that are the basis for today's powerful machine learning.John J. Hopfield created an associative memory that stores and reconstructs patterns of images and other types of data.Geoffrey Hinton invented a method that automatically discovers attributes in the data in order to perform tasks such as recognizing specific elements in an image.

When we talk about artificial intelligence, we usually refer to machine learning using artificial neural networks. This technique was originally inspired by the structure of the brain. In an artificial neural network, the neurons of the brain are represented by nodes with different values. These nodes interact with each other through connections that can be compared to synapses and can be strengthened or weakened. For example, the network is trained by creating stronger connections between nodes that also have high values. This year's winners have been doing important work on artificial neural networks since the 1980s.

John Hopfield invented a network that could use a method to preserve and reconstruct patterns. We can think of the nodes as pixels.The Hopfield network uses physics to describe the properties of a material due to its atomic spin - a property that makes each atom a tiny magnet. The entire network is described in a way that is equivalent to the energies found in spin systems in physics, and is trained by finding connection values between nodes so that the saved image has low energy. When the Hopfield network gets a distorted or incomplete image, it methodically passes through the nodes and updates their values so that the energy of the network decreases. Thus, the network gradually finds the saved image that most resembles the imperfect image it was fed.

Geoffrey Hinton used the Hopfield network as the basis for the new network, which takes a different approach - the Boltzmann machine. It learns to recognize characteristic elements in a given type of data.Hinton used the tools of statistical physics, a systems science made up of many similar elements. The machine is trained by feeding it examples that are likely to occur when it is actually running. A Boltzmann machine can be used to classify images or to create new examples for the type of pattern it was trained on. and Hinton built on this work to help unlock the current explosion in machine learning.

"The work of the winners is already delivering huge benefits. In physics, we use artificial neural networks in a wide range of areas, for example to develop new materials with specific properties." Ellen Moons, chair of the Nobel Committee for Physics, said.