English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Zendy Plus

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Zendy Tools

Zendy Open

Unsupervised graph representation learning is a non-trivial topic. Thesuccess of contrastive methods in the unsupervised representation learning onstructured data inspires similar attempts on the graph. Existing graphcontrastive learning (GCL) aims to learn the invariance across multipleaugmentation views, which renders it heavily reliant on the handcrafted graphaugmentations. However, inappropriate graph data augmentations can potentiallyjeopardize such invariance. In this paper, we show the potential hazards ofinappropriate augmentations and then propose a novel Collaborative GraphContrastive Learning framework (CGCL). This framework harnesses multiple graphencoders to observe the graph. Features observed from different encoders serveas the contrastive views in contrastive learning, which avoids inducingunstable perturbation and guarantees the invariance. To ensure thecollaboration among diverse graph encoders, we propose the concepts ofasymmetric architecture and complementary encoders as the design principle. Tofurther prove the rationality, we utilize two quantitative metrics to measurethe assembly of CGCL respectively. Extensive experiments demonstrate theadvantages of CGCL in unsupervised graph-level representation learning and thepotential of collaborative framework. The source code for reproducibility isavailable at https://github.com/zhangtia16/CGCL

CGCL: Collaborative Graph Contrastive Learning without Handcrafted Graph Data Augmentations