Domain #1: Data science and AI

Home > Areas of competence > Data science and AI

Data-driven approach to reality

The widespread deployment of communicating and precise measurement systems, coupled with efficient data storage solutions, makes the approach of external observation of complex systems realistic. Statistical learning techniques are particularly efficient. The functionalities they offer are robust, parsimonious and scalable. These techniques allow for an efficient modelling of how the systems work, whether it is for classification, detection, prediction or causality research purposes. Finally, they can be applied to many kinds of data (signals, images, videos, texts, speech, relational data, graphs, log data, dynamic data, sequences, etc.).

Challenges

Companies must take up the challenge of creating value by exploiting available data by focusing on the optimisation of internal processes and the creation of new services for their customers.

Positioning of the institute

IRT SystemX focuses on data science and artificial intelligence in its R&D projects. In industrial systems and services, data engineering and learning mechanisms provide decision support in the design and operation phases, by being integrated into processing chains. A second strand of research is the hybridization of learning solutions with knowledge models or physical models. Moreover, the institute has a specific interest in the confidence and robustness of systems integrating AI, particularly in the context of critical systems.

Roadmap

Scientific and technological challenges	Related research fields
Engineering of data and learning model	• Data and model life cycle • Model traceability/repeatability • Learning data development engineering • Knowledge acquisition, management and use • Model and algorithm embeddability
Hybridization of learning models	• With physical models • With business knowledge • With graph theory • With control theory • With multi-agent simulation • Hybrid component system
Confidence and robustness of learning models	• Interpretation of models and explanation of decisions • Characterisation of the operational domain covered by the data • Evaluation of learning models with application and functional metrics • Robust learning models
Learning in a specific context	• Federated or distributed learning • Privacy-preserving learning • Incremental learning • Representation learning • Generative models and unsupervised approaches • Transfer learning and conditions of its applicability