A Quantum Spatial Graph Convolutional Neural Network Model on Quantum Circuits

Area of Science:

• Quantum Computing
• Artificial Intelligence
• Graph Neural Networks

Background:

• Non-Euclidean data processing is a significant challenge.
• Parameterized quantum circuits (PQC) offer new computational paradigms.

Purpose of the Study:

• To propose a novel quantum spatial graph convolutional neural network (QSGCN) model.
• To enable processing of non-Euclidean data using quantum circuits.

Main Methods:

• Development of a QSGCN model with four core blocks: quantum encoding, quantum graph convolutional layer, quantum graph pooling layer, and network optimization.
• Analysis of model trainability, including the barren plateau phenomenon.
• Simulations using diverse graph datasets.

Main Results:

• Demonstration of the QSGCN model's learning capabilities on graph data.
• Validation of the model's generalization and robustness.
• Successful implementation on parameterized quantum circuit platforms.

Conclusions:

• The proposed QSGCN model is a viable approach for processing non-Euclidean data on quantum computers.
• The QSGCN model exhibits effective learning, generalization, and robustness.
• Further research into QNNs for graph data processing is warranted.