Q-Drop: Optimizing Quantum Orthogonal Networks with Statistical Pruning and Dynamic Dropout

This repository contains the source code and experimental setups for the project Q-Drop, where we investigate and develop algorithms for quantum machine learning inspired by classical dropout and pruning technique. Many quantum principles-such as superposition, entanglement, and quantum parallelism can be applied to optimize Neural Network, in this research we specifically focusing on Quantum Orthogonal Neural Network with our new proposed methods: schuduled gradients pruning and dynamic quantum dropout.

Initially, we run simulations via Pennylane and tensorflow since real quantum machine is difficult to achieve. Further, we aim to migrate and fine-tune the algorithms in order to fit the real quantum machine.

Repository Structure

Q-Drop/
├── data/*
├── notebooks/
│   ├── experiment_pruning/*
│   └── experiment_dropout/*
├── src/
│   ├── main.py
│   ├── utils/
│   │   ├── rbs_gate.py
│   │   ├── scheduled_pruning.py  
│   │   ├── dynamic_dropout.py   
│   │   └── __init__.py
│   └── models/
│       ├── orthogonal_nn.py
│       └── __init__.py
├── tmp/*
├── .gitignore
├── README.md
└── penny_env.yml

• experiment_pruning/
  Contains experimental setups and Jupyter notebooks that explore the backpropagation and the use of quantum gradient pruning technique on different datasets. Each subfolder includes tests on the following datasets:

  • MNIST (2-class)
  • Fashion MNIST (2-class)
  • Pneumonia MedMNIST
  • Retina MedMNIST (2-class)

• experiment_dropout/
  Contains the experimental setup that focuses on dynamic dropout applied to quantum orthogonal networks. Each experiment mirrors the dataset structure mentioned above.

• README.md
  Initial setup and documentation for navigating the project.

Getting Started

1. Clone the repository:

 git clone https://github.com/khanhha1005/Q-Drop-Implementation.git

2. Create conda environment:

conda env create -f penny_env.yml

3. Activate environment:

conda activate Penny2

Citation

If you use this repository or find it helpful in your research, please cite:

BibTeX

@INPROCEEDINGS{11161668,
  author={Nguyen, Pham Thai Quang and Khanh, Tran Cat and Ergu, Yared Abera and Nguyen, Van-Linh},
  booktitle={ICC 2025 - IEEE International Conference on Communications},
  title={Q-Drop: Optimizing Quantum Orthogonal Networks with Statistic Pruning and Dynamic Dropout},
  year={2025},
  pages={2394-2399},
  keywords={Training;Accuracy;Pneumonia;Neural networks;Machine learning;Stability analysis;Robustness;Optimization;Standards;Image classification;Quantum machine learning;Orthogonal neural network;Image classification;Statistical pruning;Dynamic dropout},
  doi={10.1109/ICC52391.2025.11161668}
}