Deep quanvolutional neural networks with enhanced trainability and gradient propagation.

Area of Science:

• Quantum Computing
• Artificial Intelligence
• Machine Learning

Background:

• Traditional Quanvolutional Neural Networks (QuNNs) utilize static layers, limiting feature extraction.
• Deep QuNNs struggle with gradient flow, hindering effective training and optimization.

Purpose of the Study:

• To enhance QuNN performance by introducing trainable quanvolutional layers.
• To address training challenges in deep QuNNs by improving gradient flow.

Main Methods:

• Developed trainable quanvolutional layers for enhanced feature extraction.
• Proposed Residual Quanvolutional Neural Networks (ResQuNNs) with skip connections.
• Investigated optimal placement of residual blocks for efficient training.

Main Results:

• Trainable layers significantly improved scalability and learning potential of QuNNs.
• ResQuNNs effectively enhanced gradient flow in deep networks.
• Empirical evidence confirmed strategic residual block placement optimizes training.

Conclusions:

• Trainable layers and residual learning enable effective deep learning in QuNNs.
• This work advances quantum deep learning for theoretical and practical applications.
• Findings pave the way for more powerful quantum neural network architectures.