NeurIPS 2019
Sun Dec 8th through Sat the 14th, 2019 at Vancouver Convention Center
Paper ID:395
Title:Poisson-Randomized Gamma Dynamical Systems

This is a solid contribution that proposes a novel model for sequential count tensor data, extending the existing Poisson-gamma dynamical system. The new model accounts for sparsity, and has tractable conditional distributions that enable the implementation of a Gibbs sampler. The experiments are convincing and demonstrate improvement over the PGDS and other baseline.