GraphCL: Graph Contrastive Learning with Augmentations笔记

NeurIPS 2020- Graph Contrastive Learning with Augmentations

    • contrastive learning algorithm
    • pretraining model for molecular proporty predition

GraphCL: Graph Contrastive Learning with Augmentations笔记_第1张图片
https://github.com/Shen-Lab/GraphCL

使用最基础的contrastive loss 处理图graph-level的tasks, 包括self-supervised, semi-supervised graph classification,

主要贡献是提出4种不同的augmentations.
GraphCL: Graph Contrastive Learning with Augmentations笔记_第2张图片
GraphCL: Graph Contrastive Learning with Augmentations笔记_第3张图片

contrastive learning algorithm

使用最经典的方法
GraphCL: Graph Contrastive Learning with Augmentations笔记_第4张图片
GraphCL: Graph Contrastive Learning with Augmentations笔记_第5张图片

pretraining model for molecular proporty predition

本文发现不同数据集适合不同的data 增广的方式
GraphCL: Graph Contrastive Learning with Augmentations笔记_第6张图片

你可能感兴趣的:(图神经网络,学习笔记,机器学习,深度学习)