Long Vuong

BACH: A Bayesian Admixture of Contrastive Heads for Multi-Interest Two-Tower Retrieval

Quoc Phong Nguyen, Paul Albert, Long Vuong, Vuong Le, Julien Monteil

Two-tower retrievers compress each user into a single embedding, limiting their ability to serve diverse interests. Multi-interest models give each user several heads scored by a maximum inner product, but their hard-routing training under-utilizes heads (routing collapse) and gi…Read more