Title: Probability Distributions over Joint Probability Distributions: Prediction and Empirical Validation

When: Friday, February 25, 2005, 2:00 p.m. (Talk will be of 20-30 mins) followed by question-answer session

Where: Large Common Room (5th Floor), IS Building

Who: Tomek Loboda, Information Science PhD Student

Abstract: The lognormal approximation of the distribution of the joint probability distribution proposed by Druzdzel Some properties of joint probability distributions (UAI94) has been subjected to only limited empirical study. This paper subjects this proposal to empirical tests. It shows that the lognormal approximation does not fully define the shape of the joint probability distribution distribution, and that the prediction is not accurate in the tails. Furthermore, it shows that the shape of the distribution is very robust and insensitive to influences. It additionally focuses on estimation of the parameters of that distribution, and shows, that the quality of prediction is good.

Speaker Bios
****************
Tomek Loboda is a Ph.D. student in the Department of Information Science and Telecommunications, University of Pittsburgh. He is a member of the Decision Systems Laboratory, which specializes in applying decision-theoretic techniques in decision support. His research interests include decision making under uncertainty, user interfaces and information presentation, and human factors.

E-mail: tol7@pitt.edu
Web: www.sis.pitt.edu/~tloboda