Knowledge Representation and Reasoning
In artificial intelligence, physical or virtual systems evolving in an autonomous way must be provided with capacities of representation of their environment and reasoning. The research axis “Knowledge representation and reasoning” deals with the identification, modeling, representation and implementation of the different types of information (knowledge, beliefs, preferences, actions, and so on) and setting up reasoning necessary for such systems.
Logical languages (propositional, modal, weighted, etc.) and graphical models (Bayesian networks, abstract argumentation systems, conceptual networks, etc.) are used to represent the different information (qualitative or numerical) available. The choice of the knowledge representation language is crucial to obtain good properties in terms of expressiveness, spatial efficiency, algorithmic complexity, and so on. Once the language has been chosen and the desired reasoning has been modeled, it is necessary to design practical methods to realize this reasoning, and to focus on the computational aspects of these methods, which are important in order to judge their practical use. The research carried out within the “Knowledge representation and reasoning” axis is based on the following themes and issues:
- Belief Dynamics
Development of approaches allowing the updating of an agent’s beliefs according to new information such as theory of belief change, rationality, belief updating and revision operators, logics for belief change, etc.).
Study of argumentation systems (argumentation theory, semantics for abstract argumentation, extension-based semantics, ranking-based semantics, revision and aggregation of argumentation systems, deliberation systems.
- Epistemic reasoning, reasoning about actions
Design of approaches for reasoning about action and change in the context of a dynamic multi-agent system (modal logics, epistemic logic, dynamic logic, etc.).
- Information fusion
Study of approaches for merging potentially conflicting multi-source information such as belief fusion operators, distance-based operators, preference aggregation, source reliability assessment, judgment aggregation, collective decisions.
- Conflicting information management
Development of practical methods for restoring coherence and querying from inconsistent bases that are totally or partially ordered. Development of incoherence measures, which allow to measure the extent to which several pieces of information are in conflict.
- Ontologies and description logics
Study of formal ontology formalisms based on description logics in order to take into account priorities or uncertainty in knowledge representation or revisable reasoning (reasoning in the presence of inconsistency, possibilistic logics, non-monotonic description logics, “light” description logics, ontology completion, and so on.).
- Graphical models and uncertainty
Design of formalisms based on graphical models for the representation and reasoning from uncertain or imprecise information such as Bayesian networks, possibility theory, possibilistic networks.
- Spatio-temporal reasoning
Study of formalisms allowing to reason about temporal and spatial knowledge represented qualitatively (qualitative formalisms, symbolic representation, relational languages, and so on.).
- Conceptual spaces
Development of approaches for learning representations of conceptual spaces and reasoning from conceptual spaces. Application to automatic language processing.
Development of approaches and models for knowledge compilation (knowledge compilation maps, representation languages, expressiveness, compactness, complexity). Application to product configuration.