Research


Overall objectives

COATI is a joint project-team gathering researchers from Inria, CNRS and Université Côte d’Azur. Its objectives are to conduct fundamental research in discrete mathematics, graph and digraph theory, algorithms design and operations research, and to use these knowledge and tools for addressing specific network optimization problems. Significant advances are made for instance on graph coloring problems, graph decomposition methods, combinatorial games on graphs, on the design and engineering of algorithms, etc. Furthermore, COATI addresses practical problems issued from telecommunication networks using tools from discrete mathematics and operations research in collaboration with industrial partners such as Orange labs, Nokia bell labs, Ciena, etc. We are particularly interested in optimization problems raised by the emergence of the new technologies of software defined networks (SDN) and network functions virtualization (NFV), and more specifically the placement and reconfiguration of lightpaths, network slices, service function chains, etc. We also consider the optimization of different kinds of wireless networks, including the design of reliable microwave backhaul networks, the deployment and management of fleets of drones to collect data from (mobile) sensors, and the optimization of the capacity of low power long range (LoRAWAN) networks.

During the last years, we have started to investigate how tools from artificial intelligence (AI), and in particular machine learning based methods, can help solving networks optimization problems, and how tools from (structural, metric) graph theory can help improving AI tools. More precisely, we have started to investigate the use of AI tools for networking problems, for instance for the reconfiguration of network slices in software defined networks or for the scheduling of machine learning tasks in heterogeneous clusters (Section 8.5). Furthermore, we have started to investigate the theory of deep learning, in particular by providing a rigorous understanding of some aspects of compression of artificial neural networks (Section 8.4). We have also started to investigate federated learning, for instance the privacy concern when implementing the learning algorithms in a network.

COATI also collaborates with teams in other domains (transport, biology, resource allocation, social sciences, etc.) to share its expertise for the resolution of various problems, as well as for identifying new optimization problems. Over the years, it has initiated fruitful collaborations in the fields of transport networks with SME Instant-System and Benomad (ANR Multimod 2018-2023) and with Amadeus, structural biology with project-team ABS, neurosciences with project-team CRONOS, and social sciences with SME MillionRoads and researchers from GREDEG and SKEMA.

The research done in COATI will result in the production of software components (proof of concepts) and to contributions to large open-source software such as Sagemath and packages of the Julia programming language eco-system. Finally, members of COATI are strongly involved in scientific mediation and will actively contribute to the development of Terra Numerica.

COATI has now reached the 12 years time limit for a project-team and a proposal for a new project-team (with the same name) is currently under evaluation.

Last activity report : 2024

Team members of COATI were previously members of MASCOTTE. You will find here all annual activity reports of MASCOTTE.