Shah Kalp Jayesh supervised by Dr. Subhadip Mitra received his Master of Science – Dual Degree  in Computational Natural Sciences (CND). Here’s a summary of his research work on Role of Machine Learning in Particle Physics:

 Particle physics experiments at the Large Hadron Collider (LHC) face significant challenges in distinguishing rare signal events from overwhelming background processes. While traditional cut-based methods have been foundational, they often prove inefficient for optimal signal-background classification in high-dimensional data. This thesis investigates the transformative potential of machine learning (ML) techniques—particularly Deep Neural Networks (DNNs) and Graph Neural Networks (GNNs)—to enhance classification accuracy and discovery sensitivity in Beyond Standard Model (BSM) searches. We present a phenomenological study of a leptophobic 𝑍′ boson decaying into right-handed neutrinos, employing a comprehensive simulation framework (MadGraph, Pythia, Delphes) to analyze proton-proton collisions at √𝑠 = 14 TeV. Advanced ML models are trained on kinematic variables, jet substructure observables, and physics-motivated features like invariant mass and angular correlations. Our DNN architecture achieves a 17-fold improvement in signal-to-background ratio compared to traditional methods, extending the discovery reach for 𝑍′ bosons with masses up to 5 TeV at the High-Luminosity LHC. The work highlights ML’s capacity to automate feature extraction while addressing critical challenges, including model interpretability and computational scalability. We emphasize the need for integrating explainable AI (xAI) techniques to bridge data-driven insights with physical intuition. This research establishes a framework for future studies to leverage hybrid approaches combining domain knowledge with automated learning, significantly advancing the search for new physics in the exabyte era of particle physics.

June 2025