Scalable Quantum NN

This repo is the code used to produce the results presented in "Scalable Quantum Neural Network" (SQNN).

Content

• main_qnn.py builds and trains a regular QNN that follows the circuit structure design in TensorFlow Quantum Tutorials, but uses different encoding method (angle encoding).
• main_sqnn.py builds and trains a SQNN that consists of four identical-sized quantum feature extractors and a quantum predictor. The method of data partitioning is illustrated in Fig.5(1st panel) of our paper.
• main_differsize.py builds and trains a SQNN that consists of three different-sized quantum feature extractors and a quantum predictor. The method of data partitioning is illustrated in Fig.5(3rd panel) of our paper.
• data_helper.py includes the functions for data processing.
• util.py contains the functions for recording intermediate results and quantum circuits.

How to use

Install requitements

    pip install -r requirements.txt

Set hyperparameters

To run the code for the specific tasks, please set the hyperparameters by using the command line or editing the default values of the parameters. The hyperparameters are:

Hyperparameter	Description
task	The name of the current task. A folder with the same name will be created to store intermediate results.
dataset	The name of dataset
seed	The random seed used to initialize parameters of VQCs
inputsize	The downscaled size (nxn) of input data of QNN. If the model is SQNN, the size is that before data splitting.
clfinputsize	For SQNN, the input size of qunatum predictor.
pieces	For SQNN, the number of pieces a training instance is splitted into.
lr	The learning rate
epoch	The number of training epoch
batchsize	The size of mini-batch
validation_ratio	The ratio of training data that is reserved for validation.

Run scripts

After setting the hyperparameters, run the scripts by using python3.

    python3 scale_qml/main_qnn.py 
    python3 scale_qml/main.sqnn.py
    python3 scale_qml/main_differsize.py

Check results

A folder with the same name as the task will be created after running the script to store results, and a .xls file with the same name as the task will be generated in the folder.

In the .xls file, the 1st column stores the hyperparameters of the current task; the 3rd and 4th column record the accuracy and training loss of each mini-batch; the 6th and 7th column record the accuracy and loss on the validation dataset after each training epoch; 9th and 10th column stores the accuracy and loss on the test dataset after each training epoch.

Contact

If there is any question, please send emails to jwu21@wm.edu.

Citation

If you use this code in your work, please cite our paper.

@inproceedings{wu2022wpscalable,
  title={wpScalable Quantum Neural Networks for Classification},
  author={Wu, Jindi and Tao, Zeyi and Li, Qun},
  booktitle={2022 IEEE International Conference on Quantum Computing and Engineering (QCE)},
  pages={38--48},
  year={2022},
  organization={IEEE}
}