https://arxiv.org/abs/2004.06231
Einsum Networks: Fast and Scalable Learning of Tractable Probabilistic Circuits (Robert Peharz, Steven Lang, Antonio Vergari, Karl Stelzner, Alejandro Molina, Martin Trapp, Guy Van den Broeck, Kristian Kersting, Zoubin Ghahramani)
probabilistic modeling 방법인 probabilistic circuit을 scalable하게 만든 연구. vectorize된 sum-product 모듈들을 묶어서 (레이어) einsum 한 번에 계산할 수 있게 만듦. 흥미로운 방법.
#probabilistic_model