Multi-layer belief model: systematic comparison of evidence rules of combination

Supervisor: Daira Pinto Prieto (d.pintoprieto@vu.nl)

Background

Belief function theory is a framework to combine uncertain evidence and form rational belief based on such evidence. Over the years, this theory has been extended with multiple rules to combine evidence, which ensure rationality in a given context. In (Pinto Prieto et al., 2023), the authors proposed a framework that enables the definition of combination rules based on specific parameters, which allows for adopting combination rules from other frameworks as well as creating entirely new ones. While the source discusses the theoretical advantages of this approach, an experimental and systematic comparison among the rules remains an open issue.

Description

This project aims to explore the systematic comparison of combination rules for belief functions, including traditional and new rules. Some milestones of this project are:

• Getting familiar with belief function theory and evidence modal models.
• Getting familiar with the real bird dataset with uncertain values.
• Proposing and implementing a pipeline for the systematic comparison of rules of combination.
• Extending the multi-layer belief model to enable the use of more general evidence as input.

Depending on the student’s interests, the proportion between the theoretical and the practical parts of the thesis can vary.

Literature

Daira Pinto Prieto, Ronald de Haan, and Aybüke Özgün, A Belief Model for Conflicting and Uncertain Evidence: Connecting Dempster-Shafer Theory and the Topology of Evidence. In: Proceedings of the 20th International Conference on Principles of Knowledge Representation and Reasoning (KR 2023), 2023

Constance Thierry, Arthur Hoarau, Arnaud Martin, Jean-Christophe Dubois, and Yolande Le Gall. 2022. Real Bird Dataset with Imprecise and Uncertain Values. In Belief Functions: Theory and Applications: 7th International Conference, BELIEF 2022, Paris, France, October 26–28, 2022, Proceedings. Springer-Verlag, Berlin, Heidelberg, 275–285. [PDF]