Efficient Quantum Image Classification Using Single Qubit Encoding.

Area of Science:

• Quantum Machine Learning
• Deep Learning
• Image Classification

Background:

• Deep learning models dominate image classification but suffer from high complexity and millions of parameters.
• Quantum computing offers potential solutions for computational complexity, with growing interest in quantum machine learning (QML).
• Current QML algorithms for image classification often require numerous qubits, posing challenges for the noisy intermediate-scale quantum (NISQ) era.

Purpose of the Study:

• To propose a novel single-qubit-based deep quantum neural network for image classification.
• To demonstrate the feasibility and performance of this architecture in noisy simulation environments.
• To establish a foundation for future development in efficient QML for image classification.

Main Methods:

• Developed a single-qubit deep quantum neural network architecture.
• Mimicked traditional convolutional neural network (CNN) techniques within the quantum framework.
• Conducted experiments using MNIST, Fashion-MNIST, and ORL face datasets in noisy simulation environments.

Main Results:

• Achieved classification accuracies of 94.6% on MNIST, 89.5% on Fashion-MNIST, and 82.5% on ORL face datasets.
• Demonstrated promising performance of the single-qubit architecture in noisy conditions.
• Showcased a significant reduction in parameters compared to previous QML works.

Conclusions:

• The proposed single-qubit deep quantum neural network is a viable and efficient approach for image classification in the NISQ era.
• The architecture offers a reduced parameter count, making it more feasible for current quantum hardware.
• Further research and development are warranted to enhance performance and explore broader applications.