Sites Inria

Version française

AIDA Research team

Modeling and Machine Learning for Data Interpretation and Decision Assistance

  • Leader : Jacques Nicolas
  • Research center(s) : CRI Rennes - Bretagne Atlantique
  • Field : Human-computer interaction, images processing, data management, knowledge systems
  • Theme : Databases, Knowledge Bases, Cognitive Systems

Team presentation

Aïda is a project whose general problematic is to provide an intelligent assistance to a user facing the analysis of complex and large data sets.
By "intelligent assistance", is meant the development of automatic capabilities of modeling, of recognition of interesting situations, and the ability to elaborate recommandations of suitable and comprehensible actions.
Such capabilities have an abductive or inductive nature. It means that the central issue concerns the selection of one ore several relevant hypotheses, able to best explain a given set of observations. We adopt an artificial intelligence perspective. The aim is to make the user autonomous with respect to the analysis of his data, so that the presence of a third party (specialist of data analysis, of signal processing,...) may not be required for the interpretation of the results produced by the method. Respecting this objective assumes the generation of easily interpretable results, and thus to work with conceptually simple models.

Research themes

  • Assistance for the supervision of physicals systems : We assume that fault or correct models of the monitored system are available (model-based, discrete information). As real time monitoring is expected, models are exploited off-line for chronicle acquisition or building of a diagnostic automaton called diagnoser. Only the compiled information (chronicles or diagnoser) is used on-line. We focus on dynamic systems. Therefore, modelling has to take into account a temporal dimension (temporal communicating finite state machine, temporal causal graphs). Main topics are abductive reasoning, chronicles learning, diagnoser building.
  • Machine learning of models from sequences : From a set of observations on the state or the behaviour of a system (mainly discrete sequences, that may be as well biological sequences as sequences of actions), the aim is to infer structures (finite automatas) able to explain them. For this, the project relies on works in machine learning (grammatical inference), clustering and data analysis (hierarchical clustering, decision trees).
  • Information retrieval (document) : We study the modeling of text contents through the modeling of the semantic of the elements that form their descriptors in automatic indexing. We focus on three points: development of a domain-independent model of interpretation of compounds; inference of lexical information from corpora by determining textual contexts that explain the meanings of nouns and verbs in different domains; study of the semantic variation of terms, i.e. recognition of conceptual equivalence of two structurally different sequences.

International and industrial relations

  • Project COST 15 (Multi-valued logics) and participation in the european network of excellence MONET (Model based and Qualitative Reasoning)
  • Several collaborations with Cnet (supervision of data transmission networks, syntax learning in speech recognition, belief revision of a rational agent for audiotel services)
  • Collaborations with EDF on repair planning for supply restoration in power distribution networks and on the acquisition of chronicles, Sollac, Simulog.
  • Contacts and collaborations with universities of Potsdam, Lisbon, and the Beijing mathematical institute.

Keywords: Artificial intelligence Intelligent assistance Model-based diagnosis Machine learning Information retrieval Monitoring assistance Chronicles acquisition Grammatical inference Data analysis Automatic