Assessing the Weight of Evidence Implicit in an Indefinite Probability
The authors have previously defined a novel means of quantifying uncertainty called “indefinite probabilities, designed for incorporation into the Probabilistic Logic Networks (PLN) component of the Novamente integrative AI architecture, but also with more general utility. Essentially a hybridization of Walleys imprecise probabilities and standard Bayesian credible intervals, an indefinite probability quantifies a truth value in terms of an interval L,U, a credibility level b, and an integer k called the lookahead.For comparison with other methods of uncertainty quantification, as well as for practical purposes such as indexing knowledge-items in a knowledge-base based on their evidential support, it is important to be able to convert indefinite probabilities to and from the simpler format <s,n> where s denotes a single probability estimate, and n denotes the (estimated) amount of evidence on which the estimate is based. Here we present an approach to carrying out this conversion, using an assumption of underlying Bernoulli distributions.
Ben Goertzel Matthew Iklé
Novamente LLC, 1405 Bernerd Place, Rockville, MD, 20851, USA Adams State College, Alamosa, CO, 81102 USA, and with Novamente LLC
国际会议
武汉
英文
2007-09-21（万方平台首次上网日期，不代表论文的发表时间）