Direction Reconstruction of Ultra-High Energy Cosmic Rays with Simulation-Based Inference
Sarvesh Shinde
Department of Physics & Astronomy
Faculty Supervisor: Oscar Macias
We present a simulation-based inference pipeline for reconstructing ultra-high-energy cosmic ray properties, leveraging machine learning methods guided by physical constraints. Our approach integrates Physics Informed Neural Networks (PINNs), explicitly embedding the propagation characteristics of radio waves to enhance directional reconstruction accuracy and improve uncertainty quantification. The pipeline is implemented and optimized on the SFSU GPU cluster POLARIS, enabling efficient processing and analysis of extensive training datasets. Our results demonstrate substantial improvements in modeling fidelity, emphasizing the effectiveness of physically-informed constraints in machine learning workflows for astrophysical inference.