Research Center in Artificial Intelligence

Key research subjects

Our researchers are developing innovative artificial intelligence solutions to fundamental problems of automatic natural language processing, planning, machine learning, knowledge extraction, cognitive computing, the semantic web, etc.

The scientific program of the center is organized around four main axes:

• Axis-1: Data Science.

  We pursue fundamental research on Machine Learning (ML), both deep and « shallow », Knowledge Discovery in Data (KDD), Data Mining (DM), etc. Our shared vision puts the emphasis on methods capable of accommodating heterogeneous sources of knowledge and supporting the general pattern discovery process. As a longer-term goal, we investigate the integration between components of a ML method and knowledge structures in order to enable tracing of the learning process steps and to increase its explainability, and even allow some intelligibility.

• Axis-2: Knowledge Engineering, Cognitive Architectures, Reasoning.

  A theory of cognition justifies the principles of brain function. For some years, we have been contributing to the construction of cognitive architectures implementing the most popular theories. From our perspective, the goal is to revisit the most promising architectures to found a 'cognitive plausibility' for Deep Learning (DL) processes. Rather than reducing DL to cognitively explainable models, we aim at building a common research framework enabling the development of cognitively effective ML models. In parallel, we continue our work on knowledge engineering (construction, restructuring and maintenance of ontologies, hierarchies of subjects, etc.) in a hybrid perspective, for example, by integrating ML methods where data analysis outcomes help refine representations.

• Axis-3: Automatic Planning and Decision Support.

  Automatic planning provides machines the ability to make complex decisions independently, which is essential in mobile robotics, games, and so on. At the fundamental level, we work on extending the planning models under uncertainty by the integration of reinforcement ML techniques. The recommendation uses historical data related to the consumption, evaluation or problem solving using regularity discovery, i.e. mainly ML, algorithms. Here, the benefits of incorporating domain knowledge into a recommendation engine are investigated, as well as, consumer modeling from the cognitive point of view.

• Axis-4: Automatic natural language processing (NLP).

  The NLP, as we practice it, mixes ontologies, cognitive models and statistics. Concrete subjects include machine translation, social media analysis, sentiment analysis and emotion scanning, conversational agents, and so on. Information retrieval research focuses on the discovery of structured information from texts to support the construction of knowledge graphs. On this axis, we investigate the ways to insert domain and discipline knowledge in the statistical ML techniques and the information retrieval methods of statistical inspiration. Conversely, we work on extending traditional NLP techniques of symbolic flavour with ML components to better support the construction and maintenance of large corporate knowledge structures.

---