Abstract: Topology is presented as the structural foundation of data, from molecular graphs to agent networks, enabling accurate predictions, interpretability, and robustness. The talk traces the evolution of topological learning, from empirical features and graph embeddings to graph neural networks, graph transformers, and adaptations of large language models, shifting from analyzing topology in data to embedding it in method design. The application focus is drug-drug interaction (DDI) prediction, where computational approaches are vital due to high validation costs. By modeling DDI data as graphs, topological learning supports efficient prediction. Contributions include AutoBLM, which automates discovery of scoring functions; EmerGNN, a path-aware GNN that mitigates data sparsity via pathway attention; CBR-DDI, integrating case-based reasoning with LLMs for interpretable predictions; and DDIAgent, an LLM-powered multi-agent system addressing real-world challenges. The webinar concludes by outlining future directions, such as optimizing agent collaboration, modeling LLM reasoning structures, and applying topological processes across diverse vertical domains.

Register Now

Speaker Bio: Dr. Quanming Yao, Ph.D. in Computer Science from the Hong Kong University of Science and Technology, is currently an Associate Professor in the Department of Electronic Engineering at Tsinghua University. His research focuses on the methodologies and theoretical foundations of machine learning, particularly their applications in scientific intelligence. He has long served as an Area Chair at the three major machine learning conferences: ICML, NeurIPS, and ICLR. He is also an Associate Editor for the journal Machine Learning and a Senior Editor for Neural Networks. He has received multiple internationally recognized awards, including the Early Career Spotlight Award (IJCAI 2025, AAAI 2024), InTech Prize (Ant Group, 2024), the Rising Star in Academia (AAAI, 2024), the Young Researcher Award (INNS, 2022), and the title of Rising Chinese Leader in the Global Machine Learning Field (Baidu, 2022).