ML4SCI GSoC 2026 - GENIE1 Evaluation Tasks

Evaluation tasks for the GSoC 2026 project: Deep Graph Anomaly Detection with Contrastive Learning for New Physics Searches.

Tasks

• Task 1: Autoencoder for Jet Image Representation
• Task 2: GNN Quark/Gluon Classification
• Task 3: Contrastive Learning for Classification

Data

The dataset consists of 139,306 quark and gluon jet images (3-channel, 125x125) from CMS Open Data. Raw data files are not included due to size. Run prepare_data.py to download and preprocess.py to generate preprocessed files.

Setup

pip install -r requirements.txt

Disclosure

The author used LLM tools (Claude) as a supplementary aid and independently led all aspects of the research.