GSoC 2026 - QMLHEP6

ML4Sci | Learning quantum representations of classical high energy physics data with contrastive learning

Lakshikka Sithamparanathan | Cleveland State University

---

Prerequisite Tasks

Notebook	Task
Task_I_Quantum_Circuits.ipynb	Quantum circuits + SWAP test (PennyLane)
Task_II_Classical_GNN.ipynb	GCN & GAT for quark/gluon jet classification
Task_III_Open_Task.ipynb	Commentary on quantum machine learning
Task_VI_Quantum_Contrastive_Learning.ipynb	Quantum contrastive learning with SWAP test fidelity

Key Results

Task II (50k jets, k-NN graph k=8)

Model	Accuracy	AUC
GCN (3-layer)	0.784	0.856
GAT (3-layer, 4 heads)	0.717	0.830

Task VI (MNIST, 8 qubits, 17 total wires)

Metric	Result
Same-class fidelity	0.8158 ± 0.1401
Different-class fidelity	0.6292 ± 0.1516
Fidelity gap	0.1866
Classification accuracy	77.0% @ threshold 0.667

Setup

pip install pennylane torch torchvision torch-geometric