The study of statistical features of the evolution of complex physical systems using adaptive machine learning methods

Abstract

In this work, we discuss various machine learning methods and their implementation in the field of complex physical systems for the analysis of experimental data. These methods: classical machine learning, neural nets and deep learning allow greatly outperforming classical analysis methods by giving the algorithm the ability to “learn” and perform tasks adapting to the data provided and search. Neural nets and deep learning approaches are used to search for hidden patterns of the suggested input data that can’t be analyzed using common methods. This variety of methods can be applied to study collective phenomena in plasma and thermonuclear fusion on the basis of experimental data of physical experiments with a higher level of performance than classical approaches.