图自监督学习综述

A Survey on Graph Self-supervised Learning

关键词:  自监督学习,图神经网络,图表示学习,深度学习,图分析,综述

Key words:  Self-supervised  learning, Graph  Neural Networks, Graph  Representation Learning, Deep  Learning, Graph  Analysis, Review.

摘要:

图深度学习最近引起了研究者们的兴趣。由于人工标注的高昂成本和监督学习泛化能力不足的缺点，图自监督学习（Graph Self-Supervised Learning，GSSL）通过精心设计的代理任务提取信息知识，而不依赖于人工标签，已经成为一种有前途的图数据学习范式。本文首先介绍了图自监督学习的基本概念，以及其与传统图学习方法的区别和联系；随后详细回顾了当前图自监督学习的主要方法，并将其分为图自预测、对比学习、不变挖掘和混合方法等4个大类，同时探讨和总结了图自监督学习在医学图像上的应用；最后，本文提出了未来研究的方向，包括提高算法的可解释性、使得算法适用于更大的图规模和使用算法处理结构更为复杂的异质图。

Abstract:

Graph deep learning has recently garnered interest among researchers. Given the high cost of manual annotation and the limited generalization capability of supervised learning, Graph Self-Supervised Learning (GSSL), which extracts information knowledge through carefully designed pretext tasks without relying on manual labels, has emerged as a promising and trending paradigm in graph deep learning. In this paper, we comprehensively explore the latest advancements and challenges in graph self-supervised learning. We first introduce the basic concepts of graph self-supervised learning, including its differences and connections with traditional graph learning methods. It then thoroughly reviews the current main methods of graph self-supervised learning, categorizing them into four types: graph self-prediction, contrastive learning, invariant mining, and hybrid methods. Additionally, we discuss and summarize applications of graph self-supervised learning in medical imaging. Finally, we propose future research directions, including improving algorithm interpretability and extending to more complex graph structures. Through this review, readers can gain a comprehensive understanding of the field of graph self-supervised learning, as well as the opportunities and challenges currently facing this area.