Prior feedback: Bayesian tools for maximum likelihood estimation (Q1965932)

A new method, called Prior Feedback, is proposed for the estimation of parameters of exponential families. Prior Feedback appears as an alternative algorithm for likelihood maximization. The method is based on iteration of Bayesian estimators by a series of priors. The convergence of the obtained estimator to the maximum likelihood estimator is proved when the precision of the starting prior tends to infinity. Estimates of parameters of gamma densities and parameters of a mixture of two normal distributions are examined. Estimates of parameters of logistic regression models are considered.