Graph Neural Networks (GNNs)Graph Neural Network (GNN) research at the INSIGHT Lab focuses on enhancing the expressiveness and adaptability of models for both static and dynamic (temporal) graphs.
Our researchers extend the theoretical foundations of graph learning, including advancing the Weisfeiler-Lehman (WL) theory, to substantially improve real-world benchmarks. We explore advanced aggregation methods such as Sequential Signal Mixing Aggregation (SSMA) for static graphs to increase representation power.
For dynamic graphs, we develop specialized techniques to effectively capture evolving connectivity and temporal dynamics. This research enables superior performance across applications including social network analysis, biological data, e-commerce interactions, and anomaly detection.
We explore innovative graph embeddings that seamlessly adapt to multiple tasks such as node classification, link prediction, and relational reasoning. By integrating graph structures with other modalities, we bridge gaps in complex reasoning and multimodal inference.
