talks
Invited talks, tutorials, and conference presentations by Francesco Ferrini on Graph Neural Networks, link prediction, missing features, and heterogeneous graphs.
Beyond Sparse Benchmarks: Evaluating GNNs with Realistic Missing Features
Re-evaluating progress on Graph Neural Networks under missing node features: why current sparse-feature benchmarks make every method look robust, and what changes when we move to dense, semantically meaningful features and realistic missingness mechanisms.
Meta-Path Learning for Multi-Relational Graph Neural Networks
Learning informative meta-paths in multi-relational GNNs, without handcrafted relational chains and without the scalability issues of brute-force relation weighting. A scoring function drives an incremental construction that finds the right meta-path even with many relations (e.g., knowledge graphs).
Heterogeneous Graph Learning — Hands-on Tutorial
Hands-on tutorial introducing heterogeneous graph learning: data structures, message passing on multi-relational graphs, and practical PyTorch Geometric examples for node and link prediction.