Constructing Priors in Maximum Entropy Methods
We show how to construct the best prior for a Maximum Entropy procedure when two or more priors are conceivable or are proposed. The prior is a weighed sum of the conceivable priors with weights that depend exponentially on the overlap of the prior with the exponential part of the maximum entropy probability. With additional information, one can iteratively improve the prior and sharpen the choice between alternative priors. Our construction can be used to predict in some physical cases the probability distribution functions, and to make quantitative decisions in the presence of conflicting expert opinions.
KeywordsMaximum Entropy Thought Experiment Expert Advice Maximum Entropy Method Posterior Probability Distribution
Unable to display preview. Download preview PDF.
- Englman, R., Levine, R.D. and Rivier, N. (1989) ‘On the construction of priors in maximum entropy methods’, submitted to IEEE Trans. Info. Theory; preprint, Imperial College (1987).Google Scholar
- Englman, R., Rivier, N. and Jaeger, Z. (1988 a) ‘Fragment-size distribution in disintegration by maximum-entropy formalism’, Phil. Mag.B 56, 751–769.Google Scholar
- Gull, S. (1987), seminar, Imperial College (private communication).Google Scholar
- Levine, R.D. and Tribus, M. (eds.) (1971) The Maximum Entropy Formalism, M.I.T. Press, Cambridge, MA.Google Scholar
- Mott, N.F. and Linfoot E.H. (1943), Ministry of Supply, AC 3348Google Scholar
- Renyi, A. (1970) Probability Theory, North-Holland, Amsterdam, ch. IX 5.Google Scholar
- Rivier, N. (1986) ‘Distribution of shapes and sizes by maximum entropy methods’, Ann. Isr. Phys. Soc. 8, 560–567.Google Scholar
- Rosenkrantz, R.D. (ed.) (1983) E.T. Jaynes, Papers on Probability Statistics and Statistical Physics, Reidel, Dordrecht.Google Scholar
- Stoppard, T. (1967) Rosenkrantz and Guildenstern are Dead, Faber, London.Google Scholar