Supervised exponential family principal component analysis via convex optimization

Recently, supervised dimensionality reduction has been gaining attention, owing to the realization that data labels are often available and indicate important underlying structure in the data. In this paper, we present a novel convex supervised dimensionality reduction approach based on exponential family PCA, which is able to avoid the local optima of typical EM learning. Moreover, by introducing a sample-based approximation to exponential family models, it overcomes the limitation of the prevailing Gaussian assumptions of standard PCA, and produces a kernelized formulation for nonlinear supervised dimensionality reduction. A training algorithm is then devised based on a subgradient bundle method, whose scalability can be gained using a coordinate descent procedure. The advantage of our global optimization approach is demonstrated by empirical results over both synthetic and real data.