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

This paper presents a new way to identify additional positive pairs for BYOL,a state-of-the-art (SOTA) self-supervised learning framework, to improve itsrepresentation learning ability. Unlike conventional BYOL which relies on onlyone positive pair generated by two augmented views of the same image, we arguethat information from different images with the same label can bring morediversity and variations to the target features, thus benefiting representationlearning. To identify such pairs without any label, we investigate TracIn, aninstance-based and computationally efficient influence function, for BYOLtraining. Specifically, TracIn is a gradient-based method that reveals theimpact of a training sample on a test sample in supervised learning. We extendit to the self-supervised learning setting and propose an efficient batch-wiseper-sample gradient computation method to estimate the pairwise TracIn torepresent the similarity of samples in the mini-batch during training. For eachimage, we select the most similar sample from other images as the additionalpositive and pull their features together with BYOL loss. Experimental resultson two public medical datasets (i.e., ISIC 2019 and ChestX-ray) demonstratethat the proposed method can improve the classification performance compared toother competitive baselines in both semi-supervised and transfer learningsettings.

Additional Positive Enables Better Representation Learning for Medical Images