We believe that our findings constitute a step towards establishing a hierarchy of models w.r.t. their expressive power and, in this sense, the Principal Neighbourhood Aggregation model appears to outperform the prior art in Graph Neural Networks layer design
Studying the local symmetries of graphs, we propose a more general algorithm that uses different kernels on different edges, making the network equivariant to local and global graph isomorphisms and more expressive
We hope to investigate the interpretability of the edges learned by the Graph Finite-State Automaton layer to determine whether they correspond to useful general concepts, which might allow the GFSA edges to be shared between multiple tasks
We introduce the idea of Physical Scene Graphs, which represent scenes as hierarchical graphs, with nodes in the hierarchy corresponding intuitively to object parts at different scales, and edges to physical connections between parts
Under the defined K-shot learning setting, Graph Extrapolation Networks learn to extrapolate the knowledge of a given graph to unseen entities, with a stochastic transductive layer to further propagate the knowledge between the unseen entities and model uncertainty in the link pr...
To utilize the strength of both Euclidean and hyperbolic geometries, we develop a novel Geometry Interaction Learning method for graphs, a well-suited and efficient alternative for learning abundant geometric properties in graph