Learning on Graphs
How can machine-learning models use relational structure without discarding edge types, weights, signs, layers or other information carried by a graph?
Highlights
What a generated graph should preserve
How can generated graphs preserve the structures that distinguish real network domains, and how can generation be steered towards a target class?
From brain signals to meaningful networks
How can brain activity recordings reveal the interactions that distinguish subjects and characterise brain disorders?
Research focus
Graph data are rarely homogeneous. Relations may be weighted, signed, organised in layers or observed only indirectly. We study learning methods that preserve and exploit this structure rather than flattening it before analysis.
This line includes graph neural networks for non-standard graph structures, link prediction, graph generation and methods for evaluating or interpreting learned network representations. Besides predictive performance, we look for models whose assumptions and outputs can be related back to network structure.
Selected publications
- wsGAT: Weighted and Signed Graph Attention Networks for Link Prediction, Complex Networks (2021)
- mGNN: Generalizing the Graph Neural Networks to the Multilayer Case, arXiv preprint (2021)
- Beyond MMD: Evaluating Graph Generative Models with Geometric Deep Learning, arXiv preprint (2025)
- ProtoGuide: Prototype-Driven Guidance for Class-Conditional Graph Generation, arXiv preprint (2026)