pNNCLR: Stochastic pseudo neighborhoods for contrastive learning based unsupervised representation learning problems

Download PDF
Open Webpage

Momojit Biswas, Himanshu Buckchash, Dilip K. Prasad

Published: 2024, Last Modified: 23 Sept 2025Neurocomputing 2024EveryoneRevisionsBibTeXCC BY-SA 4.0
Abstract: Highlights•Pseudo nearest neighbors improves learning in nearest neighbor contrastive methods.•A pNN based method called pNNCLR is proposed to improve upon baseline NNCLR.•Addition of noise to the sampling algorithm increases the semantic information.•A smooth-weight-updation approach is useful in controlling uncertainty in sampling.