Supervisor: Alan Bundy, +44-131-650-2716, A.Bundy@ed.ac.uk
Other Suggested Supervisors: Fiona McNeill, +44 (0) 131 650 2724, F.J.McNeill@sms.ed.ac.uk
Principal goal of the project: To build computational models of representational change driven by inference failures.
Description:
Most computational models of reasoning assume a fixed representation of knowledge, i.e. facts and rules remaining unchanged during inference. These are called 'ontologies'. However, observations of human reasoning show that representations of knowledge are fluid and evolve in response to attacks on arguments, to changing goals and to a changing world. It is not just beliefs that change, but also the 'signature' of the ontology, i.e. the functions and predicates in a logic representation, and their arities and types. My research group has pioneered techniques for identifying the need for representational change, including ontological change, and techniques for evolving the representation over time. We have applied these techniques to the semantic web, law and physics.
For her PhD, Fiona McNeill, developed ORS, a program for repairing ontologies when they lead to faulty plans. Using the initial ontology and world model, a plan is formed to achieve some goal, e.g. a travel plan. If this plan fails on execution, an analysis of the failure is carried out, which may include seeking knowledge from other agents. This analysis may suggest a repair to the world model or ontology. A replanning attempt is then made, using this new representation. The process may recurse.
This work could be extended in a number of directions.
Resources Required: Resources required depend on the precise project, but could include access to ORS and records of manual ontological change for evaluation purposes.
Degree of Difficulty: The project involves intellectual challenges about the nature of representational change, as well as requiring technical skills in knowledge representation and programming.
Background Needed: Requires knowledge of knowledge representation and reasoning, and experimental psychology. Some of this knowledge could be learnt during the project period.
Funding: There is the potential for this project to be funded by The University of Edinburgh's Neuroinformatics Doctoral Training Centre. However, note that eligibility for this funding requires UK citizenship, or long term residency in the UK for non-study purposes.
References: