Knowledge Representation and Reasonings
If we wish to conceive physical or virtual systems able to evolve autonomously, it is necessary to equip them with environment representation and reasoning capabilities.
Our research activities aim at identifying the different kinds of reasoning, model them, and implement them.
The thematic axis “Knowledge Representation and Reasoning” is dedicated to the study of different kinds of reasoning. The work concentrates mostly on the study of logical and graphical representation models of knowledge and on inference problems, belief revision, information fusion and argumentation.
Initially, one must choose a representation language to express the information. There is a vast number of possibilities: logical languages (propositional, modal, weighted, etc.), qualitative information (pre-orders, lattices, etc.), quantitative information (probabilities, utilities, etc.) or graphical models (Bayesian networks, etc.). This choice is crucial in order to obtain the desirable properties in terms of expressiveness, spacial efficiency (succinctness), algorithmic complexity, etc.
The second step is to model the different kinds of reasoning. In other words, to identify its properties in order to be able to conceive methods which perform these operations correctly.
Once the representation language is chosen and the problem is modelled, one must conceive practical methods to perform the reasoning, and then address its computational aspects. This is an important element in order to judge the possibility of implementing and using such methods in practice. This leads for instance to the study of algorithmic complexity, approximate methods, tractable fragments, compilation techniques or, when possible, the definition of practical methods giving good results despite the high complexity.