Casandra Saxon

This project aims to develop an ML-enhanced Particle Filter that uses detector response emulation to predict the reconstructed properties of truth particles during simulation. Existing particle filtering approaches rely primarily on simple kinematic cuts applied to truth particles. In this project, these cut-based selections will be replaced by integrating an ML detector-response model directly into the early stage of the simulation pipeline. Operating as a fast surrogate before full Geant4 propagation, the filter uses ML-predicted reconstructed kinematics and resolutions to dynamically evaluate particle acceptance. Particles unlikely to satisfy reconstruction criteria or those entering masked regions are aborted early, avoiding unnecessary computation while preserving physics performance. This approach complements existing techniques such as the ISF Particle Killer, which deterministically removes particles entering inactive detector regions, by providing a data-driven, detector-aware filtering strategy that can improve simulation efficiency, particularly in challenging detector transition regions.