Poisson PCA for matrix count data

We develop a dimension reduction framework for data consisting of matrices counts. Our model is based on the assumption existence small amount independent normal latent variables that drive dependency structure observed data, and can be seen as exact discrete analogue contaminated low-rank matrix model. derive estimators parameters establish their limiting normality. An extension recent proposal from literature used to estimate The method shown outperform both its vectorization-based competitors methods assuming continuity distribution in analysing simulated real world abundance data.