Studying high-energy nuclear physics with machine learning

The research paradigm in physics has evolved through three distinct phases: empirical observation and induction, theoretical modeling and deduction and computational numerical analysis and simulation. We are now situated within a novel epoch wherein the scientific research paradigm is increasingly shaped by the preeminence of large-scale data and artificial intelligence, particularly within the realm of AI for science applications. The advent of high-energy colliders coupled with Monte Carlo simulations has given rise to an unprecedented accumulation of data. Nested within this transformative research paradigm, machine learning and artificial intelligence technologies have been extensively harnessed for the analysis of these vast data sets. Within the domain of high-energy nuclear physics, two prevalent machine learning techniques have emerged: Bayesian analysis and deep learning. The former employs comprehensive fitting methodologies that compare extensive data sets against theoretical models, enabling the extraction of critical information pertaining to the initial nuclear structure, parton distributions, the equation of state governing hot and dense nuclear matter, and the transport coefficients of the quark–gluon plasma, among other parameters. Conversely, the latter capitalizes on the unparalleled pattern recognition capabilities of deep learning to discern robust features from high-dimensional raw data, specifically targeting individual physical parameters. This paper elucidates the fundamental principles of machine learning and delineates its potential to augment high-energy nuclear physics research endeavors.