Academic

Company: INRIA
Location: Rennes, France
Topic: Adaptive monitoring and model updating in a data stream context
Position type: Post-doctoral Fellow
Research theme: Cognitive systems
Project-team: DREAM - http://www.irisa.fr/activites/equipes/dream

Job description The main research topics of the research project-team DREAM (Diagnosing, REcommending Actions and Modelling) are model-based monitoring and diagnosis of time evolving systems with an emphasis on machine learning and data mining for symbolic model acquisition and construction. Application domains are medicine (cardiac monitoring), industrial diagnosis (monitoring of telecommunication networks, intrusion detection) and environmental protection.

Recently, new topics related to data stream monitoring have risen several issues, e.g. related to the fact that the monitored system evolves continuously and that the data stream contains such a huge volume of observations that it is impossible to record a complete history. Then, how to detect that the observed system has changed to such an extent that the current monitoring model is becoming obsolete? How to mine the streaming data online in order to update the model online to restore a satisfying diagnosis quality?

To tackle these issues DREAM has proposed a self-adaptive system relying on a multi-source monitoring architecture associated to meta- diagnosis. Mono-source diagnosers analyze continuously partial observations coming from the observed system. They send their diagnosis to higher level diagnosers which merge lower level diagnoses while checking their compatibility with respect to integrity constraints elaborated by experts of the domain. Whenever a constraint is violated, the meta-diagnoser identifies some diagnosers that are not consistent and the model of some of them should be updated. To this end, each faulty mono-source diagnosers mines the observed data continuously and applies its own revision procedure to update its model, when needed.

For additional information, see

www.inria.fr/travailler/opportunites/postdoc/postdoc.en.html

Qualifications The applicant should have a PhD in data mining. Skills in diagnosing and in data stream analysis would be appreciated.

_Contact_:
QUINIOU Rene mailto:quiniou@irisa.fr
INRIA / IRISA
Phone : +33 2 99 84 73 19
Campus Universitaire de Beaulieu
Fax : +33 2 99 84 71 71
35042 RENNES CEDEX - FRANCE http://www.irisa.fr/dream/site/Rene.Quiniou.html