Dynamic Filtering of Useless Data in an Adaptive Multi-Agent System : Evaluation in the Ambient Domain

Abstract

Amadeus is an Adaptive Multi-Agent System whose goal is to observe and to learn users’ behaviour in an ambient system in order to perform their recurrent actions on their behalf. Considering the large number of devices (data sources) that generally compose ambient systems, performing an efficient learning in such a domain requires filtering useless data. This paper focuses on an extended version of Amadeus taking account this requirement and proposes a solution based on cooperative interactions between the different agents composing Amadeus. An evaluation of the performances of our system as well as the encouraging obtained results are then shown.