A comparative analysis and noise robustness evaluation in quantum neural networks

Area of Science:

• Quantum Computing
• Artificial Intelligence
• Machine Learning

Background:

• Noisy Intermediate-Scale Quantum (NISQ) devices present challenges for hybrid quantum neural networks (HQNNs).
• Quantum noise significantly impacts the performance of HQNNs in practical applications.
• Image classification is a key task where HQNNs are being explored.

Purpose of the Study:

• To conduct a comparative analysis of different HQNN algorithms for image classification.
• To evaluate the performance and noise robustness of Quantum Convolutional Neural Networks (QCNN), Quanvolutional Neural Networks (QuanNN), and Quantum Transfer Learning (QTL).
• To identify optimal architectures and assess their resilience to various quantum noise channels.

Main Methods:

• Comparative analysis of QCNN, QuanNN, and QTL algorithms for image classification.
• Evaluation of algorithms across quantum circuits with varying entangling structures and layer counts.
• Assessment of noise robustness using Phase Flip, Bit Flip, Phase Damping, Amplitude Damping, and Depolarization Channel noise models.

Main Results:

• Top-performing HQNN models showed varied resilience to different quantum noise channels.
• QuanNN consistently outperformed other models, demonstrating greater robustness across various noise types.
• The selection of HQNN architecture is crucial and should consider the specific noise environment of NISQ devices.

Conclusions:

• QuanNN emerges as a more robust HQNN algorithm for image classification on NISQ devices.
• Tailoring HQNN model selection to the specific noise characteristics of quantum hardware is essential for optimal performance.
• Further research into noise mitigation strategies for HQNNs is warranted.