You are here
The research work of the DEODIS team focuses on the design, optimisation and evaluation of distributed and intelligent systems for which system complexity does not allow a formal approach to solve the problems. The team uses its expertise to provide architectural or algorithmic solutions and to evaluate their relevance, effectiveness and efficiency.
scientific objectives and thems
The scientific positioning of the DEODIS team covers modelling with different tools, the design of algorithms or protocols and optimisation, oriented towards distributed and/or intelligent systems and their their specificities. The solutions are then evaluated by simulation or in real contexts.
The team is more particularly positioned on the following research themes:
- Distributed Artificial Intelligence: We are working on the integration of tools such as case-based reasoning and deep learning to address the needs of contexts where learning data is sparse, for example to improve outcomes in medical image processing or education. Our work also focuses on efficient parallel execution (High Performance Computing) of multi-agent systems, aiming to identify emerging phenomena from individual behaviours, to enable the simulation of large-scale systems,
- Parallel scheduling: We address various contexts such as databases, with online scheduling of queries on key-value stores, data centers powered by renewable energy with scheduling constrained by the variability of electrical power, or classical problems for which we propose original solutions. We are also working on proposal of test sets to assess the quality of scheduling algorithms in parallelism or for jobshop problems.
- Security and privacy: To ensure privacy protection we are working on the anonymisation of datasets using deep learning techniques. Datasets are modified in a way that allows for a risk-free publication of records, for example in the case of medical or location data. Data analysis can also be used to the identification of patterns for classification, for example for ICD-10 codes in medicine. In the context of the protection of partially open access data, we are also interested in detection of attacks by inference.
- AI for medicine (anonymisation, security, imaging) and education
- Modelling and parallelization of multi-agent systems
- Algorithms for parallel scheduling and jobshop
- Data anonymisation
- Routing in sensor networks (work until 2021)
IMPLEMENTATING OUR WORK
The team collaborates, within the framework of projects, with numerous academic partners: Besançon University Hospital (automatic image segmentation, surveillance of the operating room), Hôpital Nord Franche-Comté (medical document processing), IRIT laboratory (scheduling under power constraints), IRD (parallelization of multi-agent systems), LIP laboratory (scheduling in key-value repositories), ... and companies to which we bring our expertise: Maincare, Aprogsys, Eaton, Orange, ...