qst-cgan

Quantum state tomography with conditional generative adversarial networks

This repository contains the full code to reproduce https://arxiv.org/abs/2008.03240 (accepted in PRL on June 10, 2021). Some parts of the figures in the paper are generated here with the implementation of the CGAN in TensorFlow. In addition, check out the implementation of accelerated-project-gradient based maximum likelihood estimation (APG-MLE) from qMLE https://arxiv.org/abs/1609.07881 to have a fast MATLAB code that reconstructs density matrices from noisy data.

Fig 1: Illustration of the CGAN architecture for QST. Data $\mathbf d$ sampled from measurements of a set of measurement operators $A$ on a quantum state is fed into both the generator $G$ and the discriminator $D$. The other input to $D$ is the generated statistics from $G$. The next to last layer of $G$ outputs a physical density matrix and the last layer computes measurement statistics using this density matrix. The discriminator compares the measurement data and the generated data for each measurement operator and outputs a probability that they match.

Installation

To run the code:

- clone this directory
- cd to the current folder `cd qst-cgan`
- make a local installation with `pip install -e .` which installs the necessary libraries such as tensorflow-cpu and qutip to run the code.
- cd to the folder paper-figures `cd paper-figures` and run the notebooks

Note

1. Please note that the code requires TensorFlow which might give you an error due to the pinned numpy version. If you face such an issue, upgrade your numpy after installing TensorFlow

pip install -U numpy

2. The Hilbert size for the output density matrices are set to 32 for now to follow the results of the paper. The last layer of the neural network needs to be edited to allow for other Hilbert space sizes. I will be putting an update on adapting this code to qubit tomography with a flexible Hilbert space size soon.

Contact

Please send me an email if you face any trouble running the code at "shahnawaz.ahmed95@gmail.com".