Should we use a uniform prior in probabilistic decision making?

• Autorzy: Piegat A. , Landowsky M.
• Języki publikacji: EN

Abstrakty

EN

In probabilistic decision-making and diagnostics problems solved with the use of Bayes' theorem and in Bayes' networks if a priori distribution of probability density is not known the uniform distribution is assumed to determine the working, approximate solution of the problem. However, in many practical problems experts possess some qualitative knowledge about a priori distribution, e.g. the distribution is unimodal (one maximum) or it is unimodal right-asymmetric one, etc. It was explained in the paper that in such cases we need not unconditionally use the uniform distribution but we may use other types of distributions which better correspond to our qualitative knowledge and to the reality. However, to determine a priori distribution compatible with our qualitative knowledge we have to calculate the approximate, average, limit distribution the AAL- distribution of an infinitely large number of all possible distributions that possess the given qualitative feature, e.g., which are unimodal and right-asymmetrical ones. Is it possible at all? In the paper it was shown that it is possible if a special method conceived by one of the authors, the method of event-granulation diminution (GD-method) is applied. With this method the Readers themselves can determine their own limit distribution of all possible distributions which possess required qualitative features. The paper gives 3 such distributions determined by one of the authors that can directly be used in problems. It was also shown that the uniform distribution itself is the average, limit distribution of an infinite number of all possible distributions. According to the authors' knowledge the presented concept of the average, safe distribution is new in the scientific literature.

Słowa kluczowe

EN

probability theory; decision-making; diagnostic problems; artificial intelligence; Bayes networks; uniform a-priori distribution

• Wydawca: Wydawnictwo Politechniki Poznańskiej
• Czasopismo: Foundations of Computing and Decision Sciences
• Rocznik: 2010
• Tom: Vol. 35, No. 1
• Strony: 43--61

Twórcy

autor

Piegat A.

autor

Landowsky M.

• West Pomeranian University of Technology, Faculty of Computer Science and Information Systems, Żołnierska 49, 71-210 Szczecin, Poland,

Bibliografia

• [1] http://en.wikipedia.org/wiki/Principle_of_indifference
• [2] http://en.wikipedia.org/wiki/Imprecise_probability
• [3] http://en.wikipedia.org/wiki/Uncertainty_reduction_theory
• [4] http://en.wikipedia.org/wiki/Bayes'_theorem
• [5] Li D., Du Y., Artificial Intelligence with Uncertainty, Chapman&Hall/CRC, Boca Raton, London, New York, 2008.
• [6] Magidar O., The classical theory of probability and the principle of indifference. 5th Annual Carnegie Mellon/University of Pittsburgh Graduate Philosophy Conference, pp. 1-17. http://www.andrew.cmu.edu/org/conference/2003
• [7] O'Hagan A., Uncertain Judgment, Eliciting Experts' Probabilities,. Willey, England, Chichester, 2006.
• [8] A. Piegat, M. Landowski, Bayes' Rule, Principle of Indifference, and Safe Distribution, proceedings of the conference AISC 2008 (Zakopane, Poland) published in "Artificial Intelligence and Soft Computing - ICAISC 2008", Springer, 2008, pp. 661-670.
• [9] A. Piegat, M. Landowski, Surmounting Information Gaps Using Average Probability Density Function, proceedings of the conference ACS 2009 (Międzyzdroje, Poland) published in "Pomiary, Automatyka, Kontrola", no. 10/2009, pp. 793-795.
• [10] Pouret O., Nairn P., Marcot B., (eds), Bayesian Networks. A practical guide to application,. John Willey & Sons Ltd, Chichester, England, 2008.
• [11] Russel R., Norvig P., Artificial Intelligence-A Modern Approach, Second edition, Prentice Hall, Upper Saddle River, 2003.
• [12] Yakov B. H., Info-gap decision theory-decisions under severe uncertainty, Second edition, Academic Press, London, 2006.