Bayesian Inference for Nonnegative Matrix Factorisation Models
Figure 1
(a) A schematic description of the NMF model with data augmentation.
(b) Graphical model with hyperparameters. Each source element is Poisson
distributed with intensity . The observations are given by . In matrix notation, we write . We can analytically integrate out over . Due to superposition property of Poisson
distribution, intensities add up, and we obtain . Given , the NMF algorithm is shown to seek the maximum
likelihood estimates of the templates and excitations . In our Bayesian treatment, we further assume that
elements of and are Gamma
distributed with hyperparameters .