A note on NPML estimation for exponential family regression models with unspecified dispersion parameter.

摘要

Nonparametric maximum likelihood (NPML)udestimation for exponential families with unspecified dispersionudparameter phi suffers from computational instability, which canudlead to highly fluctuating EM trajectories and suboptimaludsolutions, in particular when phi is allowed to vary overudmixture components. In this paper, a damped version of the EMudalgorithm is proposed to cope with these problems.