Agents, Learning and Adaptation

The goal of A ^ 3 theme is to define a set of methodologies for modeling and decision support by integrating dynamic and evolving aspects. This theme is based on our dynamical learning tools (SUSE) and our know-how on the software adaptation (ICARE) . Contributions in terms of modeling mainly cover the continuous updating of the decision spaces (adaptive classifiers based SVM and Gaussian mixtures , characteristics selection and adaptation). Concerning decision support, the main contributions refer to the construction of multi-agent simulators using dynamical modeling, the coordination (via methods of constraint satisfaction coupled with Markov Decision Processes), and adaptation (through flexible architectures and component-based agents).

Complementarity between these approaches is particularly evident through applications dedicated to the analysis and simulation of different human behaviors in interaction with complex and intelligent systems. In addition, it opens interesting perspectives about the human interfaces, robots, smart buildings.

In this theme, our goals are to develop new methods and concepts for :

  • Classification and learning of the actual behavior observed in an open environment and a changing environment,
  • Designing decision mechanisms triggering adaptation and system operation,
  • Adapt the agent software architecture in connection with the behavior change,
  • Coordinate individual agents adaptations.