B1094
Title: Computing maximum variance for interval uncertainty
Authors: Przemyslaw Grzegorzewski - Warsaw University of Technology (Poland)
Adam Kolacz - Warsaw University of Technology (Poland) [presenting]
Abstract: The specificity of interval-valued data may cause some considerable problems even at the initial step of data analysis and statistical inference. For example, it is known that in general the problem of computing sample variance under interval uncertainty (perceived from the epistemic view) is NP hard. Therefore, for practical applications one has to consider only such cases when efficient computation may be possible. Some particular classes of interval-valued data for which efficient algorithms reaching the goal in an acceptable time have been found. Unfortunately, these classes require limitations on the intervals that sometimes appear too strong for the practical use. We propose an asymptotic approach leading to a novel class of intervals for which an efficient algorithm for computing the upper endpoint of the sample variance exists (the lower endpoint can be always computed in a feasible time). Our result shows not only a broad class of intervals of interest but also characterizes this class by conditions that could be easily verified.
Journal