Meta and Google researchers’ new data curation method could transform self-supervised learning

The new technique automatically curates balanced datasets, avoiding undersampling rare data, for training self-supervised learning models.