Graph Neural Networks
My recent research has focused on Graph Neural Networks (GNNs) for learning actionable representations of graph-structured data, which is commonly generated in scientific applications. My long-term goal is to develop robust data-driven decision making tools to augment and accelerate scientific discovery, e.g. the design of novel biomolecules and sustainable materials.
Towards developing the next generation of GNN architectures, I have open-sourced better benchmarks for modern GNNs which are enabling the community to explore their theoretical and empirical limitations. A good place to get started with GNNs would be my introductory blogposts on graph representation learning with GNNs and their connection with the popular Transformer model from NLP.