Éric Villemonte de la Clergerie

Patent Representation Learning via Self-supervision

Results indicate that patent sections provide effective self-supervised positive views for learning dense patent representations, and section-pair training improves compatibility among abstracts, claims, and descriptions of the same invention.