Data-driven Learning of Geometric Scattering Networks

Alexander Tong, Frederik Wenkel, Kincaid Macdonald, Smita Krishnaswamy, Guy Wolf

Keywords: Graph Neural Networks, GNNs, Geometric Scattering, Radial Basis Network, Graph Signal Processing, Wavelet

Abstract: Many popular graph neural network (GNN) architectures, which are often considered as the current state of the art, rely on encoding graph structure via smoothness or similarity between neighbors. While this approach performs well on a surprising number of standard benchmarks, the efficacy of such models does not translate consistently to more complex domains, such as graph data in the biochemistry domain. We argue that these more complex domains require priors that encourage learning of longer range features rather than oversmoothed signals of standard GNN architectures. Here, we propose an alternative GNN architecture, based on a relaxation of recently proposed geometric scattering transforms, which consists of a cascade of graph wavelet filters. Our learned geometric scattering (LEGS) architecture adaptively tunes these wavelets and their scales to encourage band-pass features to emerge in learned representations. This results in a simplified GNN with significantly fewer learned parameters compared to competing methods. We demonstrate the predictive performance of our method on several biochemistry graph classification benchmarks, as well as the descriptive quality of its learned features in biochemical graph data exploration tasks. Our results show that the proposed LEGS network matches or outperforms popular GNNs, as well as the original geometric scattering construction, while retaining certain mathematical properties of its handcrafted (nonlearned) design.

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics

One-sentence Summary: We introduce learnable geometric scattering showing theoretical and empirical benefits in graph classification particularly in the biochemical domain.

Reviewed Version (pdf): https://openreview.net/references/pdf?id=oLeKLoB7Ns