SYNTHESIS OF PHYSICS AND DATA: MACHINE LEARNING IN RADIATION TRANSPORT PROBLEMS

Abstract

This paper reviews modern approaches to integrating Machine Learning (ML) and Artificial Intelligence (AI) methods into neutron-physical modeling tasks. The paradigm shift from purely physical models to hybrid systems is analyzed, where deep learning algorithms are used for nuclear data approximation, uncertainty quantification, and accelerating Monte Carlo convergence. Special attention is given to Physics-Informed Neural Networks (PINNs) for solving the transport equation and generating weight window maps in deep penetration problems. The issues of "black box" model interpretability and result verification in the context of nuclear safety are discussed.