This repository contains code for the paper Learning Quantum Processes with Quantum Statistical Queries. We include an implementation of our learning algorithm for predicting arbitrary quantum processes using "QPSQs". We also provide an implementation of the similar ML algorithm using classical shadows from Learning to predict arbitrary quantum processes by Huang, Chen and Preskill.
The learn method in coeff.py takes as input a unitary and an observable, and returns the learned coefficients as a dictionary. learn can be used for both HCP's classical shadow algorithm (flag = 0) and our QPSQ algorithm (flag = 1).
predict.py has a method pred that takes a dictionary of coefficients (as output by learn) and a set of states (density matrices), and returns the prediction for each state.
We generate synthetic QPSQ outputs for our simulations using the q_statistical_query method in dataset.py. We compare our synthetic data with a classical shadow method for estimating the expectation of an observable for a simple instance in simulations\obs_error.py.
We provide simulations of our learning algorithm in the simulations folder, where we simulate its performance on a range of unitaries and sets of target states. We include a similar simulation for HCP's algorithm in main.py.