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

Self-Supervised Learning (SSL) is a paradigm that leverages unlabeled datafor model training. Empirical studies show that SSL can achieve promisingperformance in distribution shift scenarios, where the downstream and trainingdistributions differ. However, the theoretical understanding of itstransferability remains limited. In this paper, we develop a theoreticalframework to analyze the transferability of self-supervised contrastivelearning, by investigating the impact of data augmentation on it. Our resultsreveal that the downstream performance of contrastive learning depends largelyon the choice of data augmentation. Moreover, we show that contrastive learningfails to learn domain-invariant features, which limits its transferability.Based on these theoretical insights, we propose a novel method calledAugmentation-robust Contrastive Learning (ArCL), which guarantees to learndomain-invariant features and can be easily integrated with existingcontrastive learning algorithms. We conduct experiments on several datasets andshow that ArCL significantly improves the transferability of contrastivelearning.

ArCL: Enhancing Contrastive Learning with Augmentation-Robust Representations