Learning multi-linear representations of distributions for efficient inference

摘要

We examine the class of multi-linear representations (MLR) for expressing probability distributions over discrete variables. Recently, MLR have been considered as intermediate representations that facilitate inference in distributions represented as graphical models.