Skip to content

Latest commit

 

History

History
60 lines (48 loc) · 3.18 KB

README.md

File metadata and controls

60 lines (48 loc) · 3.18 KB

QIANets: Quantum-Integrated Adaptive Networks for Reduced Latency and Improved Inference Times in CNN Models

Table of Contents

  1. Introduction
  2. Project Overview
  3. Research Contributions
  4. Technologies and Algorithms
  5. Performance Metrics
  6. Setup and Installation
  7. Usage
  8. Contributing
  9. Contact Information

Introduction

QIANets is a quantum-inspired approach designed to reduce latency and improve inference times in Convolutional Neural Networks (CNNs). By integrating quantum-adaptive networks with traditional CNN models like ResNet-18, GoogLeNet, and DenseNet, QIANets achieves significant model compression without sacrificing accuracy. This novel approach uses quantum-inspired techniques such as pruning, tensor decomposition, and **annealing efficient solution for real-world AI systems.

Project Overview

  • Objective: Optimize traditional CNNs by applying quantum-inspired techniques to reduce computational overhead, improve inference times, and minimize model size.
  • Key Focus:
    • Quantum-inspired methods for CNN model enhancement.
    • Model compression techniques for practical AI deployments.
    • Performance comparisons between QIANets-enhanced CNNs and baseline models.

Research Contributions

This project introduces and tests quantum-integrated methods in popular CNN architectures:

  • GoogLeNet, DenseNet, and ResNet-18: Enhanced using quantum-inspired pruning and matrix factorization techniques.
  • Latency Reduction: Demonstrates significant improvement in model inference time, enabling real-time applications.
  • Model Compression: Reduces model size by over 50%, while retaining over 90% of the original accuracy.

Published Paper:

Technologies and Algorithms

  • Quantum-Inspired Algorithms:

    • Pruning: Reduces model size by removing unnecessary parameters while maintaining model performance.
    • Tensor Decomposition: Decomposes large tensors into lower-dimensional forms for faster computation.
    • Annealing-Based Matrix Factorization: Inspired by quantum annealing, this technique optimizes weight matrices for reduced latency and enhanced efficiency.

  • CNN Architectures: This project is applied to architectures such as GoogLeNet, DenseNet, and ResNet-18.

Performance Metrics

  • Model Size Reduction: Achieves over 50% size reduction in CNN models.
  • Accuracy Retention: Maintains over 90% accuracy compared to the original models.
  • Inference Time: Reduces latency by 30% to 50%, enabling faster predictions.
  • Benchmarks: Comparative results for standard and QIANets-enhanced models across datasets like CIFAR-10 and ImageNet.

Setup and Installation

Prerequisites

  • Python 3.8+
  • TensorFlow / PyTorch
  • Quantum libraries (optional but recommended)

Installation

  1. Clone the repository: 
    git clone https://github.com/EdwardMagongo/QIANets.git
cd QIANets