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 in computer vision aims to leverage the inherentstructure and relationships within data to learn meaningful representationswithout explicit human annotation, enabling a holistic understanding of visualscenes. Robustness in vision machine learning ensures reliable and consistentperformance, enhancing generalization, adaptability, and resistance to noise,variations, and adversarial attacks. Self-supervised paradigms, namelycontrastive learning, knowledge distillation, mutual information maximization,and clustering, have been considered to have shown advances in invariantlearning representations. This work investigates the robustness of learnedrepresentations of self-supervised learning approaches focusing on distributionshifts and image corruptions in computer vision. Detailed experiments have beenconducted to study the robustness of self-supervised learning methods ondistribution shifts and image corruptions. The empirical analysis demonstratesa clear relationship between the performance of learned representations withinself-supervised paradigms and the severity of distribution shifts andcorruptions. Notably, higher levels of shifts and corruptions are found tosignificantly diminish the robustness of the learned representations. Thesefindings highlight the critical impact of distribution shifts and imagecorruptions on the performance and resilience of self-supervised learningmethods, emphasizing the need for effective strategies to mitigate theiradverse effects. The study strongly advocates for future research in the fieldof self-supervised representation learning to prioritize the key aspects ofsafety and robustness in order to ensure practical applicability. The sourcecode and results are available on GitHub.

Can Self-Supervised Representation Learning Methods Withstand Distribution Shifts and Corruptions?