5.2 Graph Contrastive Learning 对比学习的目的是最大化具有相似语义信息的实例之间的MI,因此可以构造不同的前置任务来丰富来自这些信息的监督信号。现有工作可以分为两大主流:同尺度对比学习和跨尺度对比学习。前者以相等的比例区分图实例(如节点-节点),而后者将对比放在多个粒度上(如节点-图)。图8展示了方法流程。
利用pretext任务(预训练任务),自我监督学习 (SSL) 可以从原始数据中提取有用知识,很大程度缓解了标签依赖的问题,如今已成为CV和NLP场景的主流学习范式(虽然pretext已经在CV也是昨日黄花了,但contrastive也是在这基础之上的)。但与CV和NLP等其他领域不同,图上的SSL有自己的独特场景,有不同的设计理念和分类法(指对图...
Graph Contrastive LearningCoLA [liu2021anomaly]✔✔SF✔ SL-GAD [9568697]✔✔SF✔ S: Supervised Learning, SM: Semi-supervised Learning, U: Unsupervised Learning, SF: Self-supervised Learning, O: Online Learning ApproachModelTime ComplexityDescription[b] ...
论文标题:Data Augmentation for Deep Graph Learning: A Survey 论文作者:Kaize Ding, Zhe Xu, Hanghang Tong, Huan Liu 论文来源:2022, arXiv 论文地址:download 1 介绍 本文主要总结图数据增强,并对该领域的代表性方法做出归类分析。 DGL 存在的两个问题: ...
3. 对比式学习(Contrastive Learning) 介绍完三个常见的训练策略后,我们至此完成了对图自监督相关的概念,符号等背景知识的介绍,接下来我们将逐个介绍各种方法。由于近一年来Moco [18] 和SimCLR [19] 等算法大火,各种基于互信息最大化的对比学习方法层出不穷,对比式学习的自监督方法最为大家关注和熟悉,我们也将首...
Gu Z, Luo X, Chen J, Deng M, Lai L (2023) Hierarchical graph transformer with contrastive learning for protein function prediction. Bioinformatics 39(7):btad410 Article Google Scholar Guo Z, Wang H (2020) A deep graph neural network-based mechanism for social recommendations. IEEE Trans ...
Deep learning models can accurately predict molecular properties and help making the search for potential drug candidates faster and more efficient. Many existing methods are purely data driven, focusing on exploiting the intrinsic topology and construct
2024 Propagation Tree Is Not Deep: Adaptive Graph Contrastive Learning Approach for Rumor Detection AAAI 2024 Link Link 2024 Barely Supervised Learning for Graph-Based Fraud Detection AAAI 2024 Link Link 2024 ADA-GAD: Anomaly-Denoised Autoencoders for Graph Anomaly Detection AAAI 2024 Link Link 2024...
Hinton, “A simple framework for contrastive learning of visual representations,” in International conference on machine learning. PMLR, 2020, pp. 1597–1607. [20] F. Zhuang, Z. Qi, K. Duan, D. Xi, Y. Zhu, H. Zhu, H. Xiong, and Q. He, “A comprehensive survey on transfer ...
2.1.1 Optimal Structure Learning Optimal Structure Learning 分为三类: computing node similarities via metric learning (i.e., metric-basedmethods) 代表性方法为 GAUG,AdaEdge[12],IDGL[13] optimizing adjacency matrices as learnableparameters(i.e., optimization-based methods) ...