Proven Accurate Artificial Intelligence for Nuclear and Particle Physics | MIT News
The Standard Model of particle physics describes all known elementary particles and three of the four fundamental forces governing the universe; anything but gravity. These three forces – electromagnetic, strong and weak – govern the formation of particles, their interaction and their disintegration.
The study of particle and nuclear physics in this framework is however difficult and relies on large-scale numerical studies. For example, many aspects of the strong force require numerically simulating the dynamics at the scale of 1/10th to 1/100th the size of a proton to answer fundamental questions about the properties of protons, neutrons and cores.
“Ultimately, we are computationally limited in studying proton and nuclear structure using lattice field theory,” says Phiala Shanahan, assistant professor of physics. “There are a lot of interesting problems that we know how to solve in principle, but we just don’t have enough compute, even though we’re running on the biggest supercomputers in the world.”
To overcome these limitations, Shanahan leads a group that combines theoretical physics with machine learning models. In their article “Equivariant flow-based sampling for lattice gauge theory”, published this month in Physical examination lettersthey show how integrating the symmetries of physical theories into machine learning and artificial intelligence architectures can provide much faster algorithms for theoretical physics.
“We use machine learning not to analyze large amounts of data, but to accelerate first-principles theory in a way that doesn’t compromise the rigor of the approach,” Shanahan says. “This particular work has demonstrated that we can build machine learning architectures with some of the Standard Model symmetries of particle and nuclear physics built in, and accelerate the sampling problem we are targeting by orders of magnitude.”
Shanahan started the project with MIT graduate student Gurtej Kanwar and Michael Albergo, who is now at NYU. The project has expanded to include Center for Theoretical Physics postdocs Daniel Hackett and Denis Boyda, NYU Professor Kyle Cranmer, and physics-savvy machine learning scientists from Google Deep Mind, Sébastien Racanière and Danilo. Jiménez Rezende.
This month’s article is part of a series aimed at enabling studies in theoretical physics that are currently computationally unsolvable. “Our goal is to develop new algorithms for a key component of numerical calculations in theoretical physics,” says Kanwar. “These calculations tell us about the inner workings of the Standard Model of particle physics, our most fundamental theory of matter. Such calculations are of vital importance for comparing the results of particle physics experiments, such as the Large Hadron Collider at CERN, both to constrain the model more precisely and to discover where the model breaks down and must be extended to something even more fundamental.
The only known systematically controllable method of studying the Standard Model of particle physics in the non-perturbative regime is based on sampling snapshots of quantum fluctuations in vacuum. By measuring the properties of these fluctuations, one can deduce the properties of the particles and collisions of interest.
This technique comes with challenges, says Kanwar. “This sampling is expensive, and we’re looking to use physics-inspired machine learning techniques to take samples much more efficiently,” he says. “Machine learning has already made great strides in image generation, including, for example, NVIDIA’s recent work to generate images of faces ‘imagined’ by neural networks. Considering these vacuum snapshots as images, we think it’s only natural to turn to similar methods for our problem.
Shanahan adds, “In our approach to sampling these quantum snapshots, we optimize a model that takes us from an easy-to-sample space to the target space: given a trained model, then sampling is efficient since it suffices to take independent samples. in the easy-to-sample space, and transform them via the learned pattern.
In particular, the group introduced a framework for building machine learning models that exactly respect a class of symmetries, called “gauge symmetries”, crucial for the study of high-energy physics.
As a proof of principle, Shanahan and colleagues used their framework to train machine learning models to simulate two-dimensional theory, resulting in efficiency gains of orders of magnitude over state-of-the-art techniques. and more accurate predictions from theory. This paves the way for significantly accelerating research into the fundamental forces of nature using physics-based machine learning.
The first papers of the group as a collaboration discussed the application of the machine learning technique to a simple lattice field theory and developed this class of approaches on compact and connected manifolds that describe field theories more complicated than the standard model. Now they are working to adapt the techniques to cutting-edge calculations.
“I think we’ve shown over the past year that there’s a lot of promise in combining physics insights with machine learning techniques,” Kanwar says. “We are actively considering how to overcome the remaining hurdles in how to perform large-scale simulations using our approach. I hope to see the first application of these methods to large-scale calculations within the next two years. If we are able to overcome the final hurdles, it promises to expand what we can do with limited resources, and I dream of soon performing calculations that will give us new insight into what is beyond our best understanding of physics. today.
This physics-based machine learning idea is also known to the team as “ab-initio AI”, a key theme of the Institute for Artificial Intelligence and Fundamental Interactions (IAIFI) of the National Science Foundation recently launched at MIT, where Shanahan is a researcher. coordinator for physical theory.
Led by the Nuclear Science Laboratory, IAIFI is made up of physics and artificial intelligence researchers from MIT and Harvard, Northeastern and Tufts universities.
“Our collaboration is a great example of the spirit of IAIFI, with a diverse team coming together to advance AI and physics simultaneously,” Shanahan says. Apart from research such as Shanahan’s targeting physics theory, IAIFI researchers are also working to use AI to improve the scientific potential of various facilities, including the Large Hadron Collider and the Wave Observatory. gravity of the laser interferometer, and to advance the AI itself.