Skip to main content

Reasoning with Preferences and Constraints

N. Wilson

We work on methods for expressing and reasoning with user preferences, especially on multi-variate problems, where the set of possible choices will often be restricted by constraints. A major focus of our work has been on situations where the preference relation is only partially ordered. Preference reasoning includes both (1) comparative preferences formalisms that reason directly about an agent's preference relation, and (2) formalisms that compare alternatives by computing an overall degree of satisfaction for each alternative.

Reasoning with Comparative Preferences

Comparative preference languages allow compact statements of preference over multi-variate problems. Our interests include:

  • developing expressive languages and practical inference techniques for tasks such as comparing outcomes, computing optimal outcomes, and constrained optimisation;
  • relationships with other formalisms such as lexicographic orders and weighted logics;
  • application, in particular, to recommender systems.

Soft constraints and Multi-Criteria/Objective Reasoning

In decision making problems, there is often more than one criterion (or objective) one wants to reason about. It can be hard to obtain precise tradeoffs between criteria, so this leads naturally to partially ordered preferences. In soft constraints formalisms, complete assigments are compared using the degree in which they satisfy a set of constraints. These degrees of overall satisfaction can be totally or only partially ordered. Areas of interest include the following:

  • the combination of multi-criteria and constraint-based reasoning;
  • generalised influence diagrams, that allow reasoning with multi-objective utility functions based on imprecise tradeoffs;
  • multi-objective scheduling;
  • aggregation of preferences, for example, for balancing occupant comfort in a shared office; also preference learning for such an application.