Pervasive machine learning in physics

Machine-learning-based sampling strategies have been recently developed for lattice quantum chromodynamics applications. These methods are in their early stages, but have the potential to enable currently intractable first-principles calculations in particle, nuclear and condensed matter physics and also to advance machine learning.

The rapidly developing field of physics-informed learning integrates data and mathematical models seamlessly, enabling accurate inference of realistic and high-dimensional multiphysics problems. This Review discusses the methodology and provides diverse examples and an outlook for further developments.

Machine learning methods relying on synthetic data are starting to be used in mathematics and theoretical physics. Michael R. Douglas discusses recent advances and ponders on the impact these methods will have in science.

Over the past decade machine learning has made significant advances in approximating density functionals, but whether this signals the end of human-designed functionals remains to be seen.

Recent advances in machine learning are enabling progress in several aspects of experimental fluid mechanics. This Perspective article focuses on augmenting the quality of measurement techniques, improving experimental design and enabling real-time estimation and control.

Designing new experiments in physics is a challenge for humans; therefore, computers have become a tool to expand scientists’ capabilities and to provide creative solutions. This Perspective article examines computer-inspired designs in quantum physics that led to laboratory experiments and inspired new scientific insights.

Integrated approaches with advanced machine learning techniques are becoming necessary to take full advantage of the advanced experimental capabilities of next-generation synchrotrons. Yijin Liu and colleagues discuss the emergence of synergistic machine-and-data intelligence in synchrotron technology, and how it may accelerate scientific discovery.

Graph neural networks have been applied to many important physics tasks at the Large Hadron Collider (LHC). This Technical Review categorizes these applications in a manner accessible to experts and non-experts alike by providing detailed descriptions of LHC physics and graph neural network design considerations.

Gaussian process regression (GPR) is a powerful, non-parametric and robust technique for uncertainty quantification and function approximation that can be applied to optimal and autonomous data acquisition. This Review introduces the basics of GPR and discusses several use cases from different fields.

Blending occurs when multiple sources of light occupy the same region of the sky. This Perspective discusses the problems arising from blending for astrophysical and cosmological studies, and introduces the two main strategies for solutions.

Owing to the growing volumes of data from high-energy physics experiments, modern deep learning methods are playing an increasingly important role in all aspects of data taking and analysis. This Review provides an overview of key developments, with a focus on the search for physics beyond the standard model.

Rapid advances in the capabilities of large language models and the broad accessibility of tools powered by this technology have led to both excitement and concern regarding their use in science. Four experts in artificial intelligence ethics and policy discuss potential risks and call for careful consideration and responsible usage to ensure that good scientific practices and trust in science are not compromised. 

As physicists are increasingly reliant on artificial intelligence (AI) methods in their research, we ponder the role of human beings in future scientific discoveries. Will we be guides to AI, or be guided by it?

As artificial intelligence (AI) makes increasingly impressive contributions to science, scientists increasingly want to understand how AI reaches its conclusions. Matthew D. Schwartz discusses what it means to understand AI and whether such a goal is achievable — or even needed.

Scientific understanding is one of the main aims of science. This Perspective discusses how advanced computational systems, and artificial intelligence in particular, can contribute to driving scientific understanding.

Five researchers discuss uncertainty quantification in machine-learned models with an emphasis on issues relevant to physics problems.

Artificial intelligence may uncover new scientific concepts that defy human intuition, but will researchers be able to understand and operate with them? This scenario might seem like science fiction, but physicists have faced it before.

Machine learning is no longer restricted to data analysis and is now increasingly being used in theory, experiment and simulation, that is, all traditional aspects of research. Does this perhaps signal the dawn of a new paradigm?

Finding the most appropriate machine learning algorithm for the analysis of any given scientific dataset is currently challenging, but new machine learning benchmarks for science are being developed to help.

Machine learning methods have proved powerful in particle physics, but without interpretability there is no guarantee the outcome of a learning algorithm is correct or robust. Christophe Grojean, Ayan Paul, Zhuoni Qian and Inga Strümke give an overview of how to introduce interpretability to methods commonly used in particle physics.

The number of physics articles making use of AI technologies keeps growing rapidly. There are some exciting directions worth keeping an eye on.

The rise of machine learning is moving research away from tightly controlled, theory-guided experiments towards an approach based on data-driven searches. Abbas Ourmazd describes how this change might profoundly affect our understanding and practice of physics.

Filippo Vicentini introduces the open-source Python toolkit NetKet, which implements machine learning methods for the study of quantum many body physics.

Luca Ghiringhelli introduces an AI toolkit that can be used with materials databases to discover new materials, or new properties of known materials.

Benjamin Nachman discusses the OmniFold algorithm, which combines detector simulations with machine learning to correct detector distortions.

James Spencer explains how deep neural networks can approximate many-electron wavefunctions used in variational quantum Monte Carlo, introducing the Fermionic Neural Network or FermiNet.

Johann Brehmer explains how simulation-based inference is used in particle physics and how tools such as the open-source Python library MadMiner can enhance the capabilities of data analysis.

Giulia Menichetti describes a deep learning pipeline designed to provide reliable binding predictions for poorly annotated proteins and ligands.