• PhD Student:
  • Ismahane Sid-Amar
  • Co-Advisors :
  • benferhati
  • Faiza Haned (USTHB-Alger)
  • PhD defended on :
  • Sep 20, 2015

In many applications, we are often in presence of decision making problems where the choice of appropriate actions need to be done. When the choice is clear and the risks are null, the decision becomes easy to select right actions. Decisions are more complex when available knowledge is flawed by uncertainty or when the established choice presents a risk. One of the main areas of Artificial Intelligence (AI) is to model, represent and reason about knowledge. In this thesis, we are interested in an inherent discipline in AI which concerns decision making problems.The qualitative possibility decision theory has developed several criteria, depending on the agent behavior, for helping him to make the right choice while maximizing one of these criteria. In this context, possibility theory provides a simple and natural way to encode uncertainty. It allows to express knowledge in a compact way using logical and graphical models. We propose in this thesis to study the representation and resolution of possibilistic qualitative decision problems. Possibilistic counterparts of standard approaches have been proposed and each approach aims to improve the computational complexity of computing optimal decisions and to provide more expressiveness to the representation model of the problem. In the logical framework, we proposed a new method for solving a qualitative decision problem, encoded by possibilistic bases, based on syntactic representations of data fusion problems. Subsequently, in a graphical framework, we proposed a new graphical model for decision making under uncertainty based on qualitatif possibilistic networks. Indeed, when agent’s knowledge and preferences are expressed in a qualitative way, we suggest to encode them by two distinct qualitative possibilistic networks. We developed an efficient algorithm for computing optimistic optimal decisions based on syntactic counterparts of the possibilistic networks fusion. We also showed how an influence diagram can be equivalently represented in our new model. In particular, we proposed a polynomial algorithm for equivalently decomposing a given possibilistic influence diagram into two qualitatif possibilistic networks. In the last part of the thesis, we defined the concept of negated possibilistic network that can be used for computing optimal pessimistic decisions.