Graph Generation

Graph generation focuses on creating new graphs that mimic the statistical properties of real-world graph data, aiming to generate realistic and diverse structures for various applications. Current research emphasizes diffusion models, often enhanced with techniques like flow matching and continuous-time formulations, alongside autoregressive approaches and transformer-based architectures, to improve efficiency and control over generated graph properties, including handling both discrete and continuous attributes. These advancements are significant for diverse fields, enabling applications such as drug discovery (generating novel molecules with desired properties), social network analysis, and the development of more sophisticated AI agents.