ML4SCI HEPSIM GSoC 2026 — Evaluation Task

Author: Binoy Saha | University of Texas at Arlington
Program: Google Summer of Code 2026 — ML4SCI HEPSIM

Overview

Full evaluation task solution for the ML4SCI HEPSIM GSoC 2026 projects:

• Symbolic Regression for Observable Event-Level Reweighting
• ML-Based Simulation Bias Analysis (Pythia vs Herwig)

What's in the notebook

• Part A — Data loading and exploration of the Pythia8 Quark/Gluon Jets dataset (200k jets)
• Part B — Jet observable computation: invariant mass, jet width, pT dispersion
• Part C — Lorentz boost to jet rest frame with numerical verification
• Part D — Neural network classifier (PyTorch) achieving ~79% test accuracy with ROC curve, confusion matrix, and permutation feature importance

Dataset

Pythia8 Quark and Gluon Jets — Zenodo (doi:10.5281/zenodo.3164691)
Not included in repo due to file size. Download directly from Zenodo.

Requirements

pip install numpy matplotlib scikit-learn torch jupyter