Shift Aggregate Extract Networks

TL;DR

This paper presents a deep hierarchical decomposition architecture for graph representation learning, outperforming existing methods on large social networks and being competitive on smaller datasets.

Contribution

It introduces a novel neural network framework based on hierarchical decompositions that handle high degree variability and enable domain compression for large graphs.

Findings

Outperforms state-of-the-art on large social network datasets

Competitive results on small chemobiological datasets

Efficiently handles high degree variability in graphs

Abstract

We introduce an architecture based on deep hierarchical decompositions to learn effective representations of large graphs. Our framework extends classic R-decompositions used in kernel methods, enabling nested part-of-part relations. Unlike recursive neural networks, which unroll a template on input graphs directly, we unroll a neural network template over the decomposition hierarchy, allowing us to deal with the high degree variability that typically characterize social network graphs. Deep hierarchical decompositions are also amenable to domain compression, a technique that reduces both space and time complexity by exploiting symmetries. We show empirically that our approach is able to outperform current state-of-the-art graph classification methods on large social network datasets, while at the same time being competitive on small chemobiological benchmark datasets.