Konferenční příspěvek

Proceedings of the 6th International PhD Workshop on Systems and Control

2005

RNDr. Ladislav Jirsa, Ph.D. (ÚTIA - Oddělení adaptivních systémů)

fictitious data, information matrix, LD-decomposition, Gauss

Bayesian methodology is a widely-applicable tool for probabilistic modelling and decision making. Its strength is based on a consistent way of merging information including a subjective knowledge and ideas about functioning of the uncertain modelled system. As a result, it can describe, reduce and quantify uncertainty of unknown quantities. A linear regression model with normal noise is well elaborated from this point of view. Both theoretical and stable algorithmic solutions are available. To take the full advantage of the technology, ways of building prior information and proper choice of initial information matrix representing suitable fictitious data will be discussed. The topic will be illustrated by two examples.