Heterogeneous Sheaf Neural Networks
HETSHEAF introduces a method to learn representations of heterogeneous graphs by embedding heterogeneity directly into the data structure using cellular sheaves. It learns type-aware local feature spaces and restriction maps conditioned on node, edge, and type information. SHEAFPOOL provides a universal readout mechanism that aggregates these representations in an invariant way, achieving state-of-the-art performance with high parameter efficiency.