Read the original document by opening this link in a new tab.
Table of Contents
1. Introduction
2. Overall Design
3. TF-GNN Heterogeneous Data Model
4. Modeling with TF-GNN
Summary
TF-GNN is a scalable library for Graph Neural Networks in TensorFlow designed to support rich heterogeneous graph data. The paper describes the TF-GNN data model, its Keras modeling API, and capabilities such as graph sampling, distributed training, and accelerator support. Machine learning techniques have applications across various domains, and TF-GNN addresses the need for better software frameworks for learning with graph-structured data. TF-GNN offers different levels of abstraction for increased modeling flexibility and has been designed bottom-up for modeling heterogeneous graphs. The paper discusses the layered API of TF-GNN and how it supports graph models at Google. TF-GNN enables training and inference of Graph Neural Networks on arbitrary graph-structured data. The API levels of TF-GNN allow developers of all skill levels to access powerful GNN models. The paper also details the low-level and high-level APIs used to construct GNNs with TF-GNN.