GeniePath: Graph Neural Networks with Adaptive Receptive Paths

TL;DR

GeniePath introduces an adaptive graph neural network that learns to dynamically select receptive fields, improving performance on large graph tasks by combining breadth and depth exploration of neighborhood information.

Contribution

It proposes a novel adaptive path layer for GNNs that learns importance of neighborhood sizes and filters signals from multiple hops, enhancing scalability and accuracy.

Findings

Achieves state-of-the-art results on large graphs.

Works effectively in both transductive and inductive settings.

Demonstrates scalability and improved performance over existing methods.

Abstract

We present, GeniePath, a scalable approach for learning adaptive receptive fields of neural networks defined on permutation invariant graph data. In GeniePath, we propose an adaptive path layer consists of two complementary functions designed for breadth and depth exploration respectively, where the former learns the importance of different sized neighborhoods, while the latter extracts and filters signals aggregated from neighbors of different hops away. Our method works in both transductive and inductive settings, and extensive experiments compared with competitive methods show that our approaches yield state-of-the-art results on large graphs.