QNL-Net

This repository houses the code for "A Scalable Quantum Non-local Neural Network for Image Classification"

QNL-Net Module

The core QNL-Net module is present in the qnlnet_circuit.py script, located within the relevant dataset directories.

Models

The directories mnist_digit_binaryclass and cifar10_binaryclass consists of the code for the hybrid CNN-QNL-Net and PCA-QNL-Net models for the respective datasets. These model files are named as {dataset}_model_{cnn/pca}.py.

Dependencies

Package	Version
numpy	>= 1.26.4
qiskit	>= 1.1.0
qiskit_machine_learning	>= 0.7.2
scikit_learn	>= 1.4.2
scipy	>= 1.13.1
torch	>= 2.3.0
torchsummary	>= 1.5.1
torchvision	>= 0.18.0
pandas	>= 2.1.1
matplotlib	>= 3.7.2

The dependencies for this project can be found within requirements.txt and can be installed by using pip in the command line (make sure that your pwd is the base of this repo):

pip install -r requirements.txt

Alternatively, you can simply build a virtual environment using an IDE based on this file. (recommended)

Execution

• The code files to run the models are also present within the desired dataset directories and are simply named as {dataset}_{cnn/pca}.py.
• The datasets are automatically downloaded by PyTorch when the scripts are run.
• The ansatz, feature map reps, ansatz reps, number of epochs, learning rate, and the binary classes to classify can all be configured within these scripts.
• The training data for the model runs is stored in the epoch_data subdirectory within the relevant directories mnist_digit_binaryclass and cifar10_binaryclass.

Figures & Plots

The docs directory contains figures generated programmatically for the QNL-Net ansatz and the training & accuracy plots generated using plots.ipynb.