Hybrid quantum-classical-quantum convolutional neural networks.

Area of Science:

• Quantum Computing
• Machine Learning
• Image Recognition

Background:

• Deep learning, particularly Convolutional Neural Networks (CNNs), excels at image pattern recognition.
• Hybrid quantum-classical convolutional neural networks (QCCNNs) utilize quantum properties for improved classification accuracy.
• Existing QCCNNs often lack trainable quantum parameters, limiting their learning expressivity.

Purpose of the Study:

• To propose a novel hybrid quantum-classical-quantum convolutional neural network (QCQ-CNN) architecture.
• To enhance the expressivity of decision boundaries in image classification by incorporating trainable quantum parameters.
• To evaluate the performance and robustness of QCQ-CNN on various image datasets.

Main Methods:

• Developed a QCQ-CNN integrating a quantum convolutional filter, a shallow classical CNN, and a trainable variational quantum classifier.
• Conducted small-sample experiments on MNIST, F-MNIST, and MRI tumor datasets.
• Analyzed the impact of ansatz depth and simulated quantum noise (depolarizing noise, finite sampling shots).

Main Results:

• QCQ-CNN demonstrated competitive accuracy and convergence compared to classical and hybrid baselines.
• Moderate-depth quantum circuits improved learning stability without significant complexity.
• The architecture showed a degree of robustness under simulated quantum noise conditions.

Conclusions:

• The proposed QCQ-CNN enhances expressivity in image classification through trainable quantum parameters.
• The architecture shows promise for near-term hybrid quantum learning applications, even with noise.
• Further research on larger quantum circuits and real-world quantum hardware is warranted.