Learning to learn with quantum neural networks via classical neural network (w/ PennyLane & TensorFlow)

UPDATE (04/03/2021): I'm super happy to announce that a slightly modified version of this tutorial is now featured as a demo on PennyLane's website, here: https://pennylane.ai/qml/demos/learning2learn.html

This project was created as part of the Quantum Open Source Foundation (QOSF) Mentorship Program. More on: https://qosf.org/

This repository implements the architecture proposed by Verdon et al. in the paper Learning to learn with quantum neural networks via classical neural networks [1], using PennyLane [2] and TensorFlow [3].

Project desctiption

---

Variational Quantum Algorithms (VQAs)[4] are powerful tools which promise to take full advantage of near term quantum computers. However, these algorithms suffer from optimization issues related to random initialization of the parameters. Using PennyLane and Tensorflow, this repository implements the architecture proposed by Verdon et al. in Learning to learn with quantum neural networks via classical neural networks, which leverage a classical Recurrent Neural Network (RNN) to assist the optimization of variational quantum algorithms by learning an efficient parameter initialization heuristics to ensure rapid training and convergence.

More in detail, by means of an hybrid quantum-classical recurrent setup, a Long-Short Term Memory (LSTM) is used as a black-box controller to initialize the parameters of a variational quantum circuit. In particular, in this notebook we focus our attention on the optimization of a QAOA quantum circuit to solve the MaxCut problem. An outline of the architecutre is the following, and you can find a careful explanation in the notebook.

If you need a quick recap on how to use QAOA to solve graph problems, check out this great tutorial: https://pennylane.ai/qml/demos/tutorial_qaoa_intro.html

Required packages

---

The Jupyter Notebook is written in Python, and the following packages are needed to run the code:

• PennyLane: a cross-platform Python library for differentiable programming of quantum computers, which in this tutorial will be used to create quantum circuits and interface them with
• TensorFlow: an open-source platform for machine learning, here used to build a custom model of a Recurrent Neural Network (RNN)
• NetworkX: a Python package for the creation, manipulation, and study of the structure, dynamics, and functions of complex networks
• Numpy, Matplotlib: standard libraries for array manipulation and plotting

References

---

[1] Verdon G., Broughton M., McClean J. R., Sung K. J., Babbush R., Jiang Z., Neven H. and Mohseni M. (2019),
Learning to learn with quantum neural networks via classical neural networks, arXiv:1907.05415.

[2] https://pennylane.ai/

[3] https://www.tensorflow.org/

[4] Cerezo M., Arrasmith A., Babbush R., Benjamin S. C., Endo S., Fujii K., McClean J. R., Mitarai K., Yuan X., Cincio L. and Coles P. J. (2020), Variational Quantum Algorithms, arXiv:2012.09265.

---

If you have any doubt, or wish to discuss about the project don't hesitate to contact me, I'll be very happy to help you as much as I can 😁

Have a great quantum day!