Using Machine Learning to Extract Properties of Systems of Particles

In nuclear and particle physics, one is often presented with large systems of particles in the final state that emerge from complex dynamical processes. Identifying all of these processes is challenging with traditional analytical methods. In our work, we explore the possibility of extracting information from systems of particles using machine learning algorithms trained with pseudodata from simulations. We have written Python code to create systems of particles with features such as thermal motion of particles (of a given temperature), collective motion of particles (parameterized by a flow field), and decays of particles (from unstable heavier particles of given mass). These systems resemble those seen in experimental data as a result of high energy collisions of nuclei. We apply ensemble and neural network machine learning methods to pseudodata from our simulation code in order to analyze properties like temperatures and masses of the unstable mother particles. We study the performance of various algorithms in determining these underlying parameters. If proven feasible, applications include increasing our understanding of the hadronization process and the phenomenon of confinement by analyzing experimental data with machine learning.