Collaborative and adaptive learning methods on the dynamic graphs for large networks

About the Project

Over the last 5 years, there has been a major enthusiasm for the treatment of big data, supported by a wide variety of applications. A wide range of these problems is structured in graphs and requires to adapt to varying dynamics in the time. The monitoring and management of transport, telecommunication or energy distribution are characteristic examples.
These interconnected systems consist of a large number of agents (sensors, processors, actuators) linked together by a connection topology. These agents are eventually autonomous. They can potentially interact and collaborate, dynamically, to accomplish their mission more effectively. In contrast to static databases, the flows of generated data are massive and evolve in time. The underlying graphs are themselves dynamic. Additional specificity lies in the distributed nature of the acquired data and processed by agents, encouraging distributed, collaborative and online processing of information.
ACADY aims to explore new adaptive, distributed and collaborative learning methods on very large graphs. It aims to propose contributions methodological breakthrough for extracting information from data flows at the nodes of these graphs by lifting the following locks: data size, graph size, tracking temporal and spatial variations, induced computational load.
ACADY will provide a theoretical holistic framework for the analysis and development of innovative distributed dynamic algorithms in the prospect of scaling up.
Principal Investigator
Project's partner(s)
  • Lionel FILLATRE, I3S Laboratory

October 2017 - September  2019

Total Amount

32 700 euros

Related Documentation

ACADY Presentation (Data Science Meetup, December 2018)