AbstractWe introduce ART. a new corpus-level autoencoding approach for training dense retrieval models that does not require any labeled training data. Dense retrieval is a central challenge for open-domain tasks. such as Open QA. https://jalyttlers.shop/product-category/jacket/