This is the companion repository to the paper Empirical learning of dynamical decoupling on quantum processors. You can use this code to train on physical processors then run target circuits with sequences found via GADD.
The package can be installed by running pip install . in the root directory of the package. If you would like to make changes to the package, you should run pip install -e . instead to install it in editable mode.
This package is designed to be used on top of Qiskit and the Qiskit Runtime IBM Client.
The core class GADD runs the genetic algorithm training process
on training circuits, outputting the best sequence and intermediate training data, if desired, which can be then used to run on a target circuit:
service = QiskitRuntimeService()
backend = service.least_busy(operational=True, simulator=False, min_num_qubits=n_qubits)
with Batch(backend=backend):
sampler = SamplerV2()
gadd = GADD(backend=backend)
# train
[seq, data] = gadd.train(backend=backend,
sampler=sampler,
training_circuit=training_circuit,
utility_function=utility_function,
save_iterations=True,
comparison_seqs=["baseline", "xy4", "cpmg","edd"])
# visualize the training progression
gadd.plot(seq, data)
# run on target circuit (can run on a different backend)
gadd.run(
seq = seq,
target_circuit=target_circuit,
sampler=sampler
)