Dimensionality reduction for binary data through the projection of natural parameters (Q2201560)

The main contribution of this paper is the majorization-minimization algorithm proposed for logistic PCA which makes it possible to majorize the non-convex objective function with a simpler objective and minimizing the majorizing function iteratively. The proposed logistic PCA method is applied to patient-diagnosis data. Apart from real world date, the proposed algorithm is validated on the simulated datasets from a family of multivariate Bernoulli mixtures.