CGCL: Collaborative Graph Contrastive Learning without Handcrafted Graph Data Augmentations

Tianyu Zhang,Yuxiang Ren,Wenzheng Feng,Weitao Du,Xuecang Zhang
2024-04-01
Abstract:Unsupervised graph representation learning is a non-trivial topic. The success of contrastive methods in the unsupervised representation learning on structured data inspires similar attempts on the graph. Existing graph contrastive learning (GCL) aims to learn the invariance across multiple augmentation views, which renders it heavily reliant on the handcrafted graph augmentations. However, inappropriate graph data augmentations can potentially jeopardize such invariance. In this paper, we show the potential hazards of inappropriate augmentations and then propose a novel Collaborative Graph Contrastive Learning framework (CGCL). This framework harnesses multiple graph encoders to observe the graph. Features observed from different encoders serve as the contrastive views in contrastive learning, which avoids inducing unstable perturbation and guarantees the invariance. To ensure the collaboration among diverse graph encoders, we propose the concepts of asymmetric architecture and complementary encoders as the design principle. To further prove the rationality, we utilize two quantitative metrics to measure the assembly of CGCL respectively. Extensive experiments demonstrate the advantages of CGCL in unsupervised graph-level representation learning and the potential of collaborative framework. The source code for reproducibility is available at <a class="link-external link-https" href="https://github.com/zhangtia16/CGCL" rel="external noopener nofollow">this https URL</a>
Machine Learning,Artificial Intelligence
What problem does this paper attempt to address?
### What problem does this paper attempt to solve? This paper aims to address the issue of data augmentation instability in unsupervised graph representation learning. Specifically: 1. **Problems with existing methods**: - Current Graph Contrastive Learning (GCL) methods rely on manually designed data augmentation strategies, but these augmentation strategies may lead to unstable invariance, thereby affecting model performance. - Inappropriate graph data augmentation may disrupt the structural invariance of the graph, especially in small graphs with fewer or less dense nodes. 2. **Proposed method**: - The paper proposes a new framework called "Collaborative Graph Contrastive Learning" (CGCL), which generates contrastive views through collaboration among multiple graph encoders, without relying on manually designed data augmentation. - Multiple graph encoders observe the same graph and generate different contrastive views, thus ensuring invariance and avoiding the instability and complexity brought by traditional data augmentation. 3. **Main contributions**: - A novel collaborative framework is proposed, which strengthens unsupervised graph-level representation learning by generating multiple contrastive views from the perspective of encoders. - The concept of asymmetric structure and complementary encoders is introduced as the foundational principles of the collaborative learning paradigm, and two quantitative metrics (asymmetry coefficient and complementarity coefficient) are proposed to evaluate the asymmetry and complementarity of the collaborative framework. - Experimental results show that CGCL performs excellently in graph classification tasks across nine datasets, outperforming existing methods and achieving better results without additional manually designed data augmentation. Through the above research, the authors demonstrate the advantages and potential application value of CGCL in unsupervised graph representation learning.