An UMM consists of a representation of uncertainty and a method for updating (or combining) uncertainty when new information has been obtained (e.g., from a sensor). The main interest in this work package is to evaluate and compare several different UMMs, in particular those that handle evidential information. Another problem of interest is to evaluate UMMs when used as an integral part of a system. When using such a system in a dynamic environment, it is important that the uncertainty can be traced through the reasoning process. Such tracing enables a human decision-maker to query the system in order obtain an explanation of the posterior uncertainty, e.g., that the uncertainty was mainly due to conflicting sources.
Dr. Alexander Karlsson, WP leader
Dr. Joe Steinhauer
Dr. Joeri Van Laere
Prof. Sten F Andler
Adj. Prof. George Fodor
Research Questions
The following research questions are addressed in this WP:
- Is it beneficial, with respect to decision performance, to represent lack of information (imprecision) by using evidential theories?
- Which UMMs are beneficial for obtaining a human decision-maker’s trust of a reasoning system?
- How can second-order information be utilized in order to obtain traceable uncertainty management?
Relevance to UMIF
This work package is general in nature and explores management and representation of uncertainty. Since both of these parts are an integral part of all WPs in UMIF, this package will contribute, to some extent, to all other work packages:
- WP1: Result of evaluating evidential theories and development of methods for traceable uncertainty management → (WP2-6, WP8)
- WP4, 6, 7: Results of evaluating UMMs within a specific application scenario (threat evaluation, situation detection for traffic safety and UMM for advanced driver assistance), optimization) → WP1
- WP1: Evidential software module → WP8
Collaboration with Industry
Since this is a general work package, collaboration with industry takes place through other UMIF work packages. Specifically, 2 in the former list describes collaboration with work packages that include, industrial partners: Saab, Volvo, and Volvo Technology,
Collaboration with Academia
The collaboration with other work packages is described in section “Relevance to UMIF”. In addition, there is also collaboration, regarding the topic “second-order probability”, with a researcher at the Department of Computer and Systems Sciences, Stockholm University.
Approach
The approach is to perform different type of empirical evaluations of UMMs and to develop new techniques based on these evaluations. Some of the evaluations will be conducted with respect to an application scenario, and some will be conducted by simple simulation algorithms in order to be able to analyse the behaviour and result of the UMMs more carefully.
WP Results & Status
So far we have one accepted paper:
Sundgren, D. & Karlsson, A. (2012), On Dependence in Second-Order Probability, in 'Scalable Uncertainty Management: 6th International Conference, SUM 2012 Marburg, Germany, September 17-19, 2012', pp. 379--391.
In addition, three journal papers and one conference paper is expected to be submitted before December 31, 2012.
Related Work
One example of related work is the research performed within ISIF’s ETURWG - Evaluation of Techniques for Uncertainty Representation Working Group. Some members of this WP are also a part of ETURWG. The research performed within SIPTA (The Society for Imprecise Probability: Theories and Applications) is also relevant related work.