Graph Generative Pre-trained Transformer (G2PT): An Auto-Regressive Model Designed to Learn Graph Structures through Next-Token Prediction

Researchers from Tufts University, Northeastern University, and Cornell University have developed the Graph Generative Pre-trained Transformer (G2PT), an auto-regressive model for graph generation. Unlike traditional methods, G2PT uses sequence-based representation and a transformer decoder to efficiently and accurately capture graph structures. This approach reduces computational complexity and improves scalability, making G2PT suitable for various applications, including molecular design and social network analysis. Experimental results show that G2PT performs well on diverse datasets, achieving high validity and uniqueness scores.