Deep Graph Library: A Graph-Centric, Highly-Performant Package for Graph Neural Networks

TLDR

Deep graph learning research requires new tools for efficient tensor computation over graphs. The paper introduces the design principles and implementation of the Deep Graph Library (DGL). DGL implements GNN computations as generalized sparse tensor operations, centralizes the graph abstraction for transparent optimizations, and adopts a framework‑neutral design to enable easy porting across deep learning libraries. Evaluation demonstrates that DGL outperforms other GNN frameworks in speed and memory across benchmarks, with minimal overhead on small workloads.

Abstract

Advancing research in the emerging field of deep graph learning requires new tools to support tensor computation over graphs. In this paper, we present the design principles and implementation of Deep Graph Library (DGL). DGL distills the computational patterns of GNNs into a few generalized sparse tensor operations suitable for extensive parallelization. By advocating graph as the central programming abstraction, DGL can perform optimizations transparently. By cautiously adopting a framework-neutral design, DGL allows users to easily port and leverage the existing components across multiple deep learning frameworks. Our evaluation shows that DGL significantly outperforms other popular GNN-oriented frameworks in both speed and memory consumption over a variety of benchmarks and has little overhead for small scale workloads.