Single-qubit graph classifier with classical feature aggregation

In this paper, we propose a single-qubit graph classifier that combines classical graph representation with quantum computing. Through a lightweight architecture, it achieves flexible and efficient graph data processing. This model delegates the task of aggregating node features that do not require inference to a classical subroutine, and optimizes the key weight training process with quantum programs, thereby constructing a basic graph classifier that uses only one qubit. Experimental results show that in binary classification tasks, the proposed single-qubit classifier demonstrates strong competitiveness in terms of performance when compared with traditional algorithms and quantum algorithms. Additionally, we utilize a parallel training scheme for multiple single-qubit classifiers to effectively enhance performance in multi-classification tasks. Evaluations conducted in different quantum noise simulation environments indicate that the proposed model has good robustness, and the parallel training scheme for multiple classifiers further enhances the model's robustness in multi-classification tasks. More importantly, this model can be flexibly combined with various classical graph neural networks, providing a way to promote the application of quantum graph neural networks in diverse scenarios.