A DAEM Algorithm for Mixtures of Gaussian Process Functional Regressions

Abstract

The mixture of Gaussian process functional regressions (mix-GPFR) is a powerful tool for curve clustering and prediction. Unfortunately, there generally exist a large number of local maximums for the Q-function of the conventional EM algorithm so that the conventional EM algorithm is often trapped in the local maximum. In order to overcome this problem, we propose a deterministic annealing EM (DAEM) algorithm for mix-GPFR in this paper. The experimental results on the simulated and electrical load datasets demonstrate that the DAEM algorithm outperforms the conventional EM algorithm on parameter estimation, curve clustering and prediction.