Applying machine learning to quantum control
Repo for: Neural network accelerator for quantum control, https://arxiv.org/abs/2208.02645
- Python 3.7
- TensorFlow 2.8.0
- QKeras 0.9.0
- sklearn 1.0.2
- hls4ml 0.6.0
- SciPy 1.8.1
- QuTiP 4.7.0
- Julia 1.7.2
- Juqbox.jl 0.1.30
- Environment should be created within the repository root in a folder named "juqbox_env"
The numbered files within the src folder illustrate the research process starting from model training and ending with the plotting of fidelity results.
This Python notebook trains a quantized model using QKeras.
The next Python notebook modifies the quantization of the model and then uses hls4ml to build a FPGA-targeted model. This model is queried with various gate parameters (beta) to approximate the pulse parameters (alpha).
The pulse parameters from the initial dataset (from a traditional optimizer) and from the model are used to produce gates using Juqbox.
The gate fidelity between pairs of gates is calculated. Here, the mathematically-derived golden gate, the optimizer-generated gate, and the gate resulting from our model are compared.
This notebook intakes the fidelity results and plots them.
We use the quantum control toolbox Juqbox.
- David Xu, Columbia University, d.xu@columbia.edu
- A. Baris Ozguler, Fermilab, aozguler@fnal.gov
- Giuseppe Di Gugliemo, Fermilab, gdg@fnal.gov