We use cookies to ensure that we give you the best experience on our website. You can change your cookie settings at any time. Otherwise, we'll assume you're OK to continue.

Department of Mathematical Sciences

Seminar Archives

On this page you can find information about seminars in this and previous academic years, where available on the database.

Stats4Grads: Efficient algorithms for checking consistency of probability bounds.

Presented by Nawapon Nakharutai, Department of Mathematical Sciences, Durham University

11 May 2016 13:00 in CM105

In situations where we have little data or where we have little expert opinion, instead of stating probabilities which can lead to an erroneous conclusion, we can specify probability bounds. Lower previsions (Walley, 1991) provide a good way to do this, by bounding expectation. In this study, we explore more efficient algorithms for checking an important basic consistency principle for lower previsions, called "avoiding sure loss". The problem of checking avoiding sure loss can be written as a fully degenerate linear program. This linear program can be solved by standard methods such as the simplex method or the affine scaling method. We propose a new way of reducing the size of this linear program and for minimal introduction of artificial variables. Since the simplex method can be extremely inefficient for fully degenerate linear programs, we explore whether there is a benefit in using other methods, such as the affine scaling method. We propose a simple way to obtain an initial interior solution for our problem, which is required for starting the affine scaling method. We also identify a condition under which the algorithm can detect inconsistency much earlier compared to standard stopping criteria from the literature. In future, we plan to investigate which method is the best suited for checking whether a lower prevision avoids sure loss. We hope that this work will encourage people to use more efficient algorithms for checking avoiding sure loss, instead of standard methods such as the simplex method which are potentially very inefficient for this specific problem.

Contact for more information

See the Stats4Grads page for more details about this series.