PAAI Distinguished Lecture on Generalisable and Reliable Knowledge Graph Reasoning

On 3 July 2026, the PolyU Academy for Artificial Intelligence (PAAI) and the Department of Computing held a PAAI Distinguished Lecture. Associate Professor YIN Jie from the University of Sydney shared cutting-edge research on knowledge graph reasoning.

Targeting core Knowledge Graph (KG) issues including data sparsity and dynamic knowledge evolution, the lecture covered two key innovations. First, MoEMeta, a mixture-of-experts meta-learning model, decouples shared global relational patterns from task-specific local adaptation to enable generalisation over rare relations under few-shot settings, achieving state-of-the-art performance across mainstream benchmarks. Second, the novel Conformal Path Reasoning (CPR) framework for trustworthy KGQA, adopts query-level conformal prediction to fix invalid hop-wise calibration, outputting compact answer sets with statistical validity guarantees for high-stakes fields like medical analysis. Comprehensive experiments verify CPR’s superior coverage efficiency and strong scalability on large-scale KGs with hundreds of thousands of entities.

The lecture equipped the attendees with abundant cutting-edge research ideas and establishing a premium academic platform for in-depth discussion and sharing of innovative outcomes at the intersection of knowledge adaptation, relational reasoning, and trustworthy AI.

******END******