QCG PaAC Open Project: Density Matrix Reconstruction

Track 1: Classical Shadows with Transformer Architecture

Project Overview

This project implements a machine learning model capable of reconstructing a quantum density matrix $\rho$ from measurement data (Classical Shadows). The model enforces strict physical constraints (Hermitian, Positive Semi-Definite, Unit Trace) using a Cholesky decomposition approach.

Repository Structure

• /src: Contains core source code for the model (model.py), data generation (data_gen.py), and training loops (train.py).
• /outputs: Stores saved model weights (model_weights.pt) and training logs.
• /docs: Detailed technical documentation and replication guides.

Performance Metrics

Final trained model | Loss: 0.0030 | Fidelity: 0.9872 | TraceDist: 0.0770 | Latency: 5.63ms Model saved to outputs/model_weights.pt. View metrics there

AI Attribution Policy

In compliance with the QCG PaAC Open Project guidelines:

• Tools Used: Google Gemini (https://gemini.google.com/share/82a098e65b9e)
• Usage:
  • Generated PyTorch boilerplate for the ShadowReconstructor class.
  • Debugged tensor shape mismatches in Cholesky decomposition.
• Verification:
  • Math verified against standard quantum mechanics textbooks ($\rho = LL^{\dagger}$).
  • Fidelity metric cross-referenced with standard library implementations.
  • Final results say it all. A Fidelity of 98.72% is considered really good in QST. A trace distance value of 0.077 is very low, corroborating the high fidelity score. The loss drops rapidly from 0.0817 to 0.0030 and stabilizes suggesting that the model has finished learning. A latency of 5-6 milliseconds is very fast, showing that the model can quickly recognize the states now.