Autoregressive Generative Modeling of Weighted Graphs

Richard Williams, Eric Nalisnick, Andrew Holbrook

Keywords: Autoregressive models, Graph Generative Models, Deep Neural Networks

TL;DR: This paper introduces two generative models on edge-weighted graphs

Abstract: Weighted graphs are ubiquitous throughout biology, chemistry, and the social sciences. However, most current deep generative models are designed for unweighted graphs and are not easily extended to weighted topologies. This paper proposes two autoregressive models on weighted graphs: the Adjacency-LSTM and BiGG-E. Experiments on a variety of benchmark datasets demonstrate that both models adequately capture distributions over weighted graphs while remaining computationally scalable. Specifically, we experiment with Erdős–Rényi, tree, lobster, and 3D point cloud graph data sets.

Submission Type: Extended abstract (max 4 main pages).

Submission Number: 150