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/

Global: Zendy Plus annual

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

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Unsupervised graph representation learning is critical to a wide range ofapplications where labels may be scarce or expensive to procure. Contrastivelearning (CL) is an increasingly popular paradigm for such settings and thestate-of-the-art in unsupervised visual representation learning. Recent workattributes the success of visual CL to use of task-relevant augmentations andlarge, diverse datasets. Interestingly, graph CL frameworks report strongperformance despite using orders of magnitude smaller datasets and employingdomain-agnostic graph augmentations (DAGAs). Motivated by this discrepancy, weprobe the quality of representations learnt by popular graph CL frameworksusing DAGAs. We find that DAGAs can destroy task-relevant information and harmthe model's ability to learn discriminative representations. On small benchmarkdatasets, we show the inductive bias of graph neural networks can significantlycompensate for this weak discriminability. Based on our findings, we proposeseveral sanity checks that enable practitioners to quickly assess the qualityof their model's learned representations. We further propose a broad strategyfor designing task-aware augmentations that are amenable to graph CL anddemonstrate its efficacy on two large-scale, complex graph applications. Forexample, in graph-based document classification, we show task-awareaugmentations improve accuracy up to 20%.

Augmentations in Graph Contrastive Learning: Current Methodological Flaws & Towards Better Practices