2026-05-06T09:35:32.530Z http://ezaposleni.singidunum.ac.rs/rest/sciNaucniRezultati/oai

ezaposleni.singidunum.ac.rs/rest/sciNaucniRezultati/oai:1:2527 2014-06-11T06:27:18Z 1

A new adaptive robustified prediction algorithm with unknown noise statistics 2014 http://ezaposleni.singidunum.ac.rs/rest/sciNaucniRezultati/oai/record/1/2527 http://etran.etf.rs/index_e.html D. Đurović J. Gavrilović I. Kostić-Kovačević B. Kovačević A new robust adaptive predictor for situations where noise statistics are not fully known is presented in the paper. First an optimized predictor is developed, based on the minimization of a generalized mean square prediction error. It determines the structure of the robust adaptive predictor, which is synthesized through minimization of a modified criterion in which a quadratic function is introduced instead of an arbitrary non-linear function of the prediction error. The non-linear function is determined by applying Huber’s min-max approach, which assumes a priori knowledge of the distribution class to which the actual unknown noise distribution belongs. The resulting non-linearity is a maximum likelihood criterion function, and is determined by the least favorable probability density function within the given class, which carries minimal information about the estimated parameters. Unknown parameters of the predictor are estimated at each step by applying a recursive algorithm of the stochastic gradient type conferenceObject Conference proceedings IcETRAN