Probing the Quark Gluon Plasma Using Machine Learning

The Quark Gluon Plasma (QGP) is a state of matter, thought to be akin to the conditions of the early universe, which cannot be directly observed. Jets can be used as indirect probe to study the properties of the QGP. In this talk, we present a study to probe the differences in jet quenching under different conditions based on experimentally observable quantities or quantities that can be derived from experimental results. These physical quantities are generated using a variety of simulations including Pythia8, Hydjet++ and Jewel as validation for future data-driven studies. Although machine learning techniques are used in high-energy physics, this work builds on existing frameworks to develop techniques optimized for physics-based predictive tasks in heavy-ion collisions. This work considers gradient boosted random forest models where variables used for analysis can be correlated directly with the response (decision). Further, the training inputs, whether or not correlated, can be ranked according to the contribution they have to the final decision, allowing for the different subtleties underlying the decision to be captured. This transparency allows for different types of physics effects to be attributed to the decision, making this a promising tool for doing analysis in actual experiments.