Code for the paper Sample-efficient Model-based Reinforcement Learning for Quantum Control.
To setup, run on the command line:
pip install -e .
To train a model-based policy optimization agent, an example command to train the model on the 2-qubit transmon would look like:
python3 mbsac.py --epochs 2000 --system "transmon" --verbose True --experiment_name "transmon_model_no_shots_1" --buffer_size 2000 --respawn True --use_rl_model True --use_ham_model True --use_totally_random_ham True
The model-based policy optimization algorithm in the repo that was introduced in When to Trust Your Model: Model-Based Policy Optimization is based upon the starter code from this repo.
If you use this code, please cite the following paper:
@article{PhysRevResearch.5.043002,
title = {Sample-efficient model-based reinforcement learning for quantum control},
author = {Khalid, Irtaza and Weidner, Carrie A. and Jonckheere, Edmond A. and Schirmer, Sophie G. and Langbein, Frank C.},
journal = {Phys. Rev. Res.},
volume = {5},
issue = {4},
pages = {043002},
numpages = {21},
year = {2023},
publisher = {American Physical Society},
doi = {10.1103/PhysRevResearch.5.043002},
url = {https://link.aps.org/doi/10.1103/PhysRevResearch.5.043002}
}