Convergence Behavior of the em algorithm for the multivariate t -distribution |
| |
Authors: | Olcay Arslan Patrick DL Constable John T Kent |
| |
Institution: | Department of Statistics , University of Leeds , LEEDS LS2 9JT, UK |
| |
Abstract: | Iterative reweighting (IR) is a popular method for computing M-estimates of location and scatter in multivariate robust estimation. When the objective function comes from a scale mixture of normal distributions the iterative reweighting algorithm can be identified as an EM algorithm. The purpose of this paper is to show that in the special case of the multivariate t-distribution, substantial improvements to the convergence rate can be obtained by modifying the EM algorithm. |
| |
Keywords: | Convergence Rate EM Algorithm Interative Reweighting Algorithm Multivariate t-distribution Robustness |
|