2025 Data Science Project

General Description

The core idea of the project is to use this paper as a baseline.
The paper describes how Gaussian elimination can be used to synthesize a Clifford tableau into a quantum circuit.
In each step of the Gaussian elimination, a pivot row and column need to be chosen, which affects the performance of the algorithm.
The overall goal of the project is to select a pivot such that the number of CNOT gates, which are considered more costly on quantum devices, is minimized.

The paper is currently implemented in the PauliOpt library.
The algorithm described in the paper can be executed as follows:

def our_compilation(circuit: Circuit, topology: Topology, n_rep: int):
    """
    :param circuit: Quantum circuit to process
    :param topology: Device topology information
    :param n_rep: Number of repetitions for the algorithm
    :return: Processed circuit data and metadata
    """
    clifford_tableau = CliffordTableau(circuit.n_qubits)
    clifford_tableau = tableau_from_circuit(clifford_tableau, circuit)

    circ_out, _ = synthesize_tableau_perm_row_col(clifford_tableau, topology)
    return {"n_rep": n_rep} | collect_circuit_data(circ_out) | {"method": "normal_heuristic"}

Core Idea

We have adapted the library to allow for a callback that selects the pivot row and column:

perm_iter = iter(permutation) # <- some permutation that was user choosen

def pick_pivot_callback(G, remaining: "CliffordTableau", remaining_rows: list[int], choice_fn=min):
    return next(perm_iter)

qc, _ = synthesize_tableau_perm_row_col(clifford_tableau, topo, pick_pivot_callback=pick_pivot_callback)

Your task is to find good heuristics using machine learning techniques.

Setup

The project uses pyenv combined with pip-tools.
Follow the instructions in their repository to install it.

To create a new environment, run:

make clean_setup

To install dependencies, run:

make setup

To update dependencies (run pip-compile), run:

make update_dependencies

There is currently no linter setup. If you want to use linting, feel free to implement it.

Project Structure

• nn/: Contains code to generate a dataset for the neural network-based approach.
• rl/: Includes the reinforcement learning (RL)-based agent and its environment.
• Top-level Python modules:
  • nn_train_main.py: A script to generate datasets.
  • nn_eval_main.py: A script for evaluating the neural network approach.
  • rl_main.py: A script to run the RL algorithm.