Implementation of high-performance graph neural network distributed learning framework

• Main Author: Lee, Cheng Han
• Other Authors: Luo Siqiang
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2023
• Subjects: Engineering::Computer science and engineering::Computing methodologies::Artificial intelligence
• Online Access: https://hdl.handle.net/10356/166564
• Institution: Nanyang Technological University

Graph Neural Network (GNN), which uses a neural network architecture to effectively learn information organized in graphs with nodes and edges, has been a popular topic in deep learning research in recent years. Generally, distributed deep learning uses multiple devices to collaboratively train a global model with relatively low cost and high efficiency. Implementing distributed learning approaches to train GNNs is a promising and challenging task. Compared to traditional distributed learning, distributively training GNNs requires the topology of graph structures to be considered, with the utilization of graph algorithms including graph clustering and partitioning. The goal of this project is to build a distributed framework for training GNNs, and apply graph algorithms to improve learning performance, that is, to make the learning process more efficient and scalable in distributed environments. The project contains research on the current algorithms for high-performance deep learning and development of the framework based on the currently available tools in distributed learning and GNN training.