On Joint Prediction-Detection Systems,
Abstract
Several schemes are proposed for predicting future values of a random process when insufficient a priori information is available for minimum mean-squared error prediction. In the first scheme, it is assumed that the process is adequately modelled by one of M models. A system is constructed which comprises M predictors, each designed to be appropriate for a specific model, and a detector for deciding which predictor is to be used. A prediction inhibitor is incorporated in the detector for deciding which predictor is to be used. A prediction inhibitor is incorporated in the detector, so that no prediction need be made when the probability of an incorrect detector decision is high. The joint Bayesian optimization of the individual components is performed, and the system's structure and performance are discussed. The second scheme, appropriate when very little a priori information is available, presents linear minimax predictors for signals observed in white noise, given only a bound on the absolute value of their k-th derivatives. (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- Aug 01, 1972
- Accession Number
- AD0748200
Entities
People
- David William Kelsey
Organizations
- University of Illinois Urbana–Champaign