AI研讨会 | Integrating Relational Learning in Foundation Models（Rex Ying）
Despite many generations of algorithms and deep learning models for relational data, existing state-of-the-art methods, even including foundation models, have not converged to a satisfactory solution towards universal, general and practical graph intelligence. As one of the most general yet underexplored modalities, problems in the form of graphs, relational data and geometry pose some of the ultimate challenges in deep learning. This talk explores reasons why such machine learning problems are so challenging, and potential directions where foundation models can interact with graph data towards this goal. We proposed models that integrates graphs with foundation models to synergistically solve problems that require reasoning in multiple modalities, demonstrating examples where foundation model approaches improves graph learning; and where graph structure enhances LLM reliability. Finally, these new challenging tasks help redefine the problem settings and use cases towards more realistic and meaningful applications of graph and geometry, therefore necessitate new benchmarks and evaluation for future research. We demonstrate this via downstream large-scale applications of literature foundation models and domain-specific AI agents.
分享者简介
Rex Ying is an assistant professor in the Department of Computer Science at Yale University. His research focus includes geometric deep learning, foundation models with structured data, multimodal models, AI for science, and trustworthy deep learning. He is interested in the use of graphs and geometry to enhance representation learning in expressiveness and trustworthiness, in large-scale settings. Rex has built multi-modal foundation models in engineering, natural science, social science and financial domains. He won the best dissertation award at KDD 2022, and the Amazon Research Award in 2024.
顶部