Spectral Networks and Locally Connected Networks on Graphs

Citations Over Time

Abstract

Convolutional Neural Networks are extremely efficient architectures in image and audio recognition tasks, thanks to their ability to exploit the local translational invariance of signal classes over their domain. In this paper we consider possible generalizations of CNNs to signals defined on more general domains without the action of a translation group. In particular, we propose two constructions, one based upon a hierarchical clustering of the domain, and another based on the spectrum of the graph Laplacian. We show through experiments that for low-dimensional graphs it is possible to learn convolutional layers with a number of parameters independent of the input size, resulting in efficient deep architectures.

Related Papers

• → Convolutional Neural Networks On Graphs With Fast Localized Spectral Filtering (Gdl Seminar)(2016)5,066 cited
• → Deep Convolutional Networks on Graph-Structured Data(2015)1,346 cited