Skip to yearly menu bar Skip to main content


Poster

SpeqNets: Sparsity-aware permutation-equivariant graph networks

Christopher Morris · Gaurav Rattan · Sandra Kiefer · Siamak Ravanbakhsh

Hall E #412

Keywords: [ T: Deep Learning ] [ DL: Graph Neural Networks ]


Abstract: While message-passing graph neural networks have clear limitations in approximating permutation-equivariant functions over graphs or general relational data, more expressive, higher-order graph neural networks do not scale to large graphs. They either operate on $k$-order tensors or consider all $k$-node subgraphs, implying an exponential dependence on $k$ in memory requirements, and do not adapt to the sparsity of the graph. By introducing new heuristics for the graph isomorphism problem, we devise a class of universal, permutation-equivariant graph networks, which, unlike previous architectures, offer a fine-grained control between expressivity and scalability and adapt to the sparsity of the graph. These architectures lead to vastly reduced computation times compared to standard higher-order graph networks in the supervised node- and graph-level classification and regression regime while significantly improving standard graph neural network and graph kernel architectures in terms of predictive performance.

Chat is not available.