QIANets: Quantum-Integrated Adaptive Networks

QIANets is a cutting-edge model compression framework leveraging quantum-inspired techniques to reduce the size and inference time of deep learning models without sacrificing accuracy. By integrating concepts such as quantum pruning, tensor decomposition, and annealing-based matrix factorization, QIANets achieves highly efficient model compression for convolutional neural networks (CNNs), tested on GoogLeNet, ResNet-18, and DenseNet architectures.

Key Features

• Quantum-Inspired Pruning: Selectively removes non-essential model weights inspired by quantum measurement principles.
• Tensor Decomposition: Breaks down large weight matrices for more compact representations, inspired by quantum state factorization.
• Annealing-Based Matrix Factorization: Optimizes compression via quantum-inspired annealing, achieving higher compression without degrading model performance.
• Low Latency & High Efficiency: Reduces inference times on CNNs while maintaining comparable accuracy to the original model.

Table of Contents

• Overview
• Features
• Getting Started
• Installation
• Usage
• Quantum-Inspired Techniques
• Results
• Contributing
• License
• Contact

Getting Started

Prerequisites

Before running this project, ensure you have the following:

• Python 3.x
• TensorFlow or PyTorch
• Basic knowledge of quantum computing (helpful, but not mandatory)

Installation

1. Clone the repository:

   git clone https://github.com/edwardmagongo/Quantum-Inspired-Model-Compression
   cd QIANets

2. Install dependencies:

   pip install -r requirements.txt

3. Set up your environment for quantum-inspired computations:

   • [Optional] Install quantum computing libraries such as Qiskit for deeper exploration of the quantum principles.

Usage

1. Train a base CNN model (e.g., ResNet-18) using the provided dataset:

   python train.py --dataset <dataset> --model <model-type>

2. Compress the model using quantum-inspired techniques:

   python compress.py --model <trained-model-path> --compression-rate <rate>

3. Evaluate the compressed model:

   python evaluate.py --model <compressed-model-path> --dataset <dataset>

Example Workflow

1. Train a model:

   python train.py --dataset cifar10 --model resnet18

2. Compress using a 75% rate:

   python compress.py --model models/resnet18.h5 --compression-rate 0.75

3. Evaluate:

   python evaluate.py --model models/resnet18_compressed.h5 --dataset cifar10

Quantum-Inspired Techniques

1. Quantum-Inspired Pruning: We draw from quantum measurement theory to prune unimportant weights based on probabilistic outcomes, reducing model size while maintaining fidelity.
2. Tensor Decomposition: Inspired by the decomposition of quantum states, this technique factorizes large weight matrices into smaller, efficient components.
3. Annealing-Based Matrix Factorization: Employs a quantum-inspired annealing process to optimize the factorization, balancing accuracy and compression efficiency.

Results

In extensive testing on CNN models such as GoogLeNet and DenseNet, QIANets achieved:

• 50-70% reduction in inference times
• Compression rates of up to 80% without significant loss in accuracy
• Faster deployment of models on resource-constrained devices (e.g., mobile phones)

The approach shows significant promise for edge AI applications and large-scale deployment in real-time systems.

Contributing

We welcome contributions! To get involved:

1. Fork the repo and create a new branch:

   git checkout -b feature/your-feature

2. Commit your changes:

   git commit -m "Add your feature"

3. Push your branch:

   git push origin feature/your-feature

4. Open a Pull Request for review.

License

This project is licensed under the MIT License. See the LICENSE file for details.

Contact

For questions or collaboration opportunities, reach out to:

Edward Magongo
Email: [email protected]

---

Thank you for your interest in QIANets!