Closed oceanlvr closed 2 years ago
冒昧打扰前辈,不好意思。 SimGCL上的工作很有启发性,但是我有点疑问。文章提出关键不是droupout增强,而是均匀的分布达到去偏的效果 但题目起的是《Are Graph Augmentations Necessary? Simple Graph Contrastive Learning for Recommendation》让我觉得有一些迷惑,dourpout base的增广在 GBRSs 上是不太好的,其他非 droupout 增广方法在 GBRs 上是否有必要?
你可以看下我的综述https://arxiv.org/abs/2203.15876 里面总结了不同的data augmentation的类型,基于结构的扰动只是一种,还有其他种类。
谢谢前辈百忙之中的回复,谢谢 :)
别太客气~ 交流学习
冒昧打扰前辈,不好意思。 SimGCL上的工作很有启发性,但是我有点疑问。文章提出关键不是droupout增强,而是均匀的分布达到去偏的效果 但题目起的是《Are Graph Augmentations Necessary? Simple Graph Contrastive Learning for Recommendation》让我觉得有一些迷惑,dourpout base的增广在 GBRSs 上是不太好的,其他非 droupout 增广方法在 GBRs 上是否有必要?