Abstract:Graph contrastive learning (GCL) is a promising direction toward alleviating the label dependence, poor generalization and weak robustness of graph neural networks, learning representations with invariance, and discriminability by solving pretasks. The pretasks are mainly built on mutual information estimation, which requires data augmentation to construct positive samples with similar semantics to learn invariant signals and negative samples with dissimilar semantics to empower representation discriminability. However, an appropriate data augmentation configuration depends heavily on lots of empirical trials such as choosing the compositions of data augmentation techniques and the corresponding hyperparameter settings. We propose an augmentation-free GCL method, invariant-discriminative GCL (iGCL), that does not intrinsically require negative samples. iGCL designs the invariant-discriminative loss (ID loss) to learn invariant and discriminative representations. On the one hand, ID loss learns invariant signals by directly minimizing the mean square error (MSE) between the target samples and positive samples in the representation space. On the other hand, ID loss ensures that the representations are discriminative by an orthonormal constraint forcing the different dimensions of representations to be independent of each other. This prevents representations from collapsing to a point or subspace. Our theoretical analysis explains the effectiveness of ID loss from the perspectives of the redundancy reduction criterion, canonical correlation analysis (CCA), and information bottleneck (IB) principle. The experimental results demonstrate that iGCL outperforms all baselines on five node classification benchmark datasets. iGCL also shows superior performance for different label ratios and is capable of resisting graph attacks, which indicates that iGCL has excellent generalization and robustness. The source code is available at https://github.com/lehaifeng/ T-GCN/tree/master/iGCL.

Graph Contrastive Representation Learning with Input-Aware and Cluster-Aware Regularization

Debiased Graph Contrastive Learning.

Graph Representation Learning via Contrasting Cluster Assignments

Local Structure-aware Graph Contrastive Representation Learning

Contrastive Graph Representation Learning with Adversarial Cross-view Reconstruction and Information Bottleneck

Graph Contrastive Learning with Node-Level Accurate Difference

Relation-aware Graph Contrastive Learning

Graph Clustering with High-Order Contrastive Learning

Dual Contrastive Learning Network for Graph Clustering

SMGCL: Semi-supervised Multi-view Graph Contrastive Learning

InfoGCL: Information-Aware Graph Contrastive Learning

Graph Fine-Grained Contrastive Representation Learning

Adversarial Cluster-Level and Global-Level Graph Contrastive Learning for node representation

CSGCL: Community-Strength-Enhanced Graph Contrastive Learning

Dual-Channel Graph Contrastive Learning for Self-Supervised Graph-Level Representation Learning

Graph Soft-Contrastive Learning via Neighborhood Ranking

CGCL: Collaborative Graph Contrastive Learning without Handcrafted Graph Data Augmentations

Graph Contrastive Learning with Cross-view Reconstruction

Subgraph Networks Based Contrastive Learning

Unbiased and augmentation-free self-supervised graph representation learning

Augmentation-Free Graph Contrastive Learning of Invariant-Discriminative Representations