Imprecision - Lecture Day at Durham: 20 June 2006

The Department of Mathematical Sciences, Durham University, organizes a lecture day on `Imprecision', on Tuesday 20 June 2006, with topics including imprecise probability and related statistical inference, imprecise utility, info-gap theory and imprecise metrics. Recent research results will be presented, and there will be ample opportunity for discussions. Below you will find the time schedule and the abstract of the talks. If you want more information, contact me . Everyone is welcome to attend, if you wish to join the speakers for lunch please let me know by noon on Monday 19 June.


Details

The talks will take place on Tuesday 20 June 2006 in lecture room CM107 in the Department of Mathematical Sciences, Durham. This is number 15 on the map. More information about travelling to Durham can be found here.


Schedule
 

Time

Speaker

Institution

Title (click on the title to obtain the corresponding abstract)

10.00-10.45am Jim Hall University of Newcastle upon Tyne, UK Uncertainty representation in climate change projections and engineering adaption decisions using imprecise probabilities
10.45-11.15am Martin Tunnicliffe Kingston University London, UK The use of statistical induction for in-service QoS monitoring of real-time network connections
11.15-11.45am Daniel Hine University of Newcastle upon Tyne, UK Robust decision making under severe uncertainty - an information-gap approach
11.45-14.00pm Lunch
14.00-14.45pm Gert de Cooman Ghent University, Belgium Representation invariance in immediate prediction
14.45-15.15pm Frank Coolen Durham University, UK Nonparametric predictive inference for multinomial data
15.15-15.45pm Tea
15.45-16.30pm Thomas Augustin Ludwig-Maximilians University Munich, Germany Decision making with imprecise probabilities - some results and many questions
16.30-17.00pm Michael Goldstein Durham University, UK Bayes linear imprecision
17.00pm End

Abstracts

Thomas Augustin: `Decision making with imprecise probabilities -- some results and many questions'

The talk will review some basic concepts and recent computational advances in decision making with imprecise probabilites. Then it will turn to current challenges including statistical applications and dynamical (in)coherence in sequential decision making.

Gert de Cooman: `Representation invariance in immediate prediction'

Consider an exchangeable (finite or infinite) sequence of random variables. An uncertainty model for predicting the value of the n+1-th variable based on observations of the previous n variables, is called a predictive system. It will usually depend on the set of possible values -- the category space -- for these variables. When it doesn't, we call the predictive system representation invariant. The talk touches upon general properties of representation invariant predictive systems, and discusses the relationships with Walley's Imprecise Dirichlet (Multinomial) Model, (ID(M)M). I argue that, in some specific sense, and under some additional assumptions, the ID(M)M leads to the most conservative representation invariant predictive systems. The talk then goes on to identify important open problems, as well as peculiar and intriguing properties of the ID(M)M.

Jim Hall: `Uncertainty representation in climate change projections and engineering adaptation decisions using imprecise probabilities'

Whilst the majority of the climate research community is now set upon the objective of generating probabilistic predictions of climate change, disconcerting reservations persist. Attempts to construct probability distributions over socio-economic scenarios are doggedly resisted. Variation between published probability distributions of climate sensitivity attests to incomplete knowledge of the prior distributions of critical parameters and structural uncertainties in climate models. In this talk we address these concerns by adopting an imprecise probability approach. We demonstrate the approach in an imprecise probability analysis of the increase in global mean temperature. The uncertainty analysis is continued at a scale more recognisable to civil engineers by examining decision-making in relation to flood alleviation. We demonstrate the use of info-gap theory for decision-making under severe epistemic uncertainties and provide examples of the use of imprecise probabilities in the analysis of options for upgrading London's tidal flood defences.

Martin Tunnicliffe: `The use of statistical induction for in-service QoS monitoring of real-time network connections'

Traditional network Quality-of-Service (QoS) architectures typically tag timing/sequencing information to packet headers, obtaining precise loss/delay metrics at the expense of bandwidth. The author investigates inductive methods whereby imprecise metrics for an entire connection (or aggregation) can be obtained from a minority of tagged packets, thus conserving more bandwidth for payload data.

Daniel Hine: `Robust decision making under severe uncertainty - an information-gap approach'

Information-gap decision theory, developed by Yakov Ben-Haim, provides a novel framework for decision making in cases of severe uncertainty. Uncertainty is represented by families of nested convex sets. This is justified by the phenomena of convex clustering, avoiding the need to fit any form of measure or membership function. This makes the method particularly valuable in cases where extreme events are of particular interest. An overview of the principles behind the method will be presented, followed by a simple illustrative example of a flood management decision.

Michael Goldstein: `Bayes linear imprecision'

The Bayes linear approach to statistics is built on the notion of expectation, rather than probability, as the primitive concept. Within this formulation, belief specification and analysis may be based on a relatively small collection of underlying quantifications, and so allows the incorporation of imprecision in a relatively natural and straightforward way. We will discuss how this can be done, and how such analysis may be naturally extended to considerations of imprecise utility.

Frank Coolen: `Nonparametric predictive inference for multinomial data'

An update will be given on our research on this topic (first presented at ISIPTA'05), presenting an interval probabilistic method for inference that is an alternative to Walley's Imprecise Dirichlet Model. Particular attention will be given to the difference between situations with known or unknown number of possible data categories. (This is joint work with Thomas Augustin)


Frank Coolen.


Last revision: 01/06/06