kdm
Kernel density matrices provide a simpler yet effective mechanism for representing joint probability distributions of both continuous and discrete random variables. This abstraction allows the construction of differentiable models for density estimation, inference, and sampling, and enables their integration into end-to-end deep neural models.