Accurate Image Multi-Class Classification Neural Network Model with Quantum Entanglement Approach.

Area of Science:

• Quantum computing and artificial intelligence intersection.
• Development of novel quantum machine learning algorithms.

Background:

• Quantum machine learning (QML) is a rapidly growing field.
• Previous models like Quanvolutional Neural Networks (QuanvNN) have shown promise in image classification.
• Existing QML methods often require complex parameter optimization within quantum circuits.

Purpose of the Study:

• To evaluate and enhance image classification accuracy using QML.
• To introduce a new QML model, Neural Network with Quantum Entanglement (NNQE), that minimizes quantum circuit parameter optimization.
• To assess the suitability of the proposed QML model for noisy intermediate-scale quantum (NISQ) computers.

Main Methods:

• Implemented a Quanvolutional Neural Network (QuanvNN) with random quantum circuits for image classification on MNIST and CIFAR-10 datasets.
• Proposed and implemented a Neural Network with Quantum Entanglement (NNQE) model utilizing strongly entangled quantum circuits and Hadamard gates.
• Compared the performance of QuanvNN and NNQE against traditional neural networks and evaluated their efficacy on the German Traffic Sign Recognition Benchmark (GTSRB) dataset.

Main Results:

• QuanvNN improved accuracy on MNIST from 92.0% to 93.0% and CIFAR-10 from 30.5% to 34.9%.
• NNQE further boosted accuracy on MNIST to 93.8% and CIFAR-10 to 36.0%.
• NNQE demonstrated reduced reliance on quantum circuit parameter optimization and showed potential for NISQ devices.
• Performance degraded on the more complex GTSRB dataset, decreasing accuracy from 82.2% to 73.4%.

Conclusions:

• The proposed NNQE model offers improved image classification accuracy on benchmark datasets with limited quantum resources.
• The method's independence from quantum circuit parameter optimization makes it suitable for current quantum hardware.
• Further research is needed to understand the factors influencing performance improvements and degradations, especially for complex, colored datasets.