CTQWformer: A CTQW-based Transformer for Graph Classification

Researchers have introduced CTQWformer, a novel hybrid graph learning framework that integrates continuous-time quantum walks (CTQW) with Graph Neural Networks (GNNs) to enhance graph classification tasks. Addressing the limitations of existing GNN and Transformer architectures in capturing global structural dependencies and dynamic information propagation, this model employs a trainable Hamiltonian to fuse graph topology with node features. This approach enables physically grounded modeling of quantum walk dynamics, extracting rich structural information. The framework incorporates these representations into two complementary modules: a Graph Transformer module that uses final-time propagation probabilities as structural biases in self-attention, and a Graph Recurrent Module that captures temporal evolution patterns via bidirectional recurrent networks. Extensive experiments on benchmark datasets demonstrate that CTQWformer outperforms traditional graph kernel and GNN-based methods. As the first hybrid CTQW-based Transformer, it highlights the potential of integrating quantum dynamics into trainable deep learning frameworks, marking a significant advancement in graph representation learning and artificial intelligence research.