This is the code for "Community-Invariant Graph Contrastive Learning" (CI-GCL). CI-GCL adopt learnable data augmentation with Community-Invariant constraint on both topology and features. And all these parts are jointly optimized to make sure the augmentation schemes can benefit from contrastive loss, CI constraints and downstream classifiers.

• We propose a learnable CI-GCL framework to automat-ically maintain CI during graph augmentation by max-imizing spectral change loss, improving the model’s downstream performances. • We theoretically show that the proposed CI constraint can be applied to both topology and feature augmenta-tion, enhancing the model’s robustness.

Based on our observations, we propose a community-invariant GCL framework to maintain graph community structure during learnable graph augmentation. By maximizing the spectral changes, this framework unifies the constraints of both topology and feature augmentation, enhancing the model’s robustness.

2024年5月9日 · 本文介绍了一种名为“社区不变图对比学习”（ci-gcl）的新框架，旨在解决当前图对比学习（gcl）方法中存在的泛化能力有限和对噪声敏感的问题。 当前GCL方法主要依赖随机图增强，这可能会破坏重要的图信息，例如社区结构。

Based on our observations, we propose a community-invariant GCL framework to maintain graph community structure during learnable graph augmentation. By maximizing the spectral changes, this framework unifies the constraints of both topology and feature augmentation, enhancing the model's robustness.

2013年3月10日 · This is the code for "Community-Invariant Graph Contrastive Learning" (CI-GCL). CI-GCL adopt learnable data augmentation with Community-Invariant constraint on both topology and features. And all these parts are jointly optimized to make sure the augmentation schemes can benefit from contrastive loss, CI constraints and downstream classifiers.

2024年5月2日 · Based on our observations, we propose a community-invariant GCL framework to maintain graph community structure during learnable graph augmentation. By maximizing the spectral changes, this framework unifies the constraints of both topology and feature augmentation, enhancing the model's robustness.

2024年5月2日 · The key technical contribution of this research is the development of a community-invariant GCL (CI-GCL) framework for learnable graph augmentation. The framework is designed to maintain the graph community structure during the augmentation process, which is crucial for learning well-generalized node/graph representations.

from GCL.utils import (compute_infonce, cluster_get, CustomDataLoader, compute_cluster_constrain_loss,

现有的大多数图对比学习（graph contrastive learning，GCL）方法首先对输入图进行随机扩充，得到两个视角的图，通过模型学习 图嵌入表示 来最大化两个视图中表示的一致性。在 graph augmentation 过程中大多数方法采用 uniform data augmentation schemes 可能破坏原有的内在图 ...