A post-doctoral position is available at Université Côte d’Azur!
« Online/Incremental Learning for Spatial Applications »
Artificial Intelligence, Online learning, Incremental/Continual Learning, Influence functions, data pruning, federated learning
Executing machine learning (ML) algorithms on satellite could offer many advantages. It will indeed reduce the required bandwidth, the volume of of data to store on-board as well as the system responsiveness. Nowadays, ML algorithms are still trained on earth before being deployed on the satellite. Once on-board, these algorithms face different challenges related to data variabilities (light conditions, sensors ageing, etc.). Training these models again is therefore required. To do so, we will first study approaches that aim at minimizing the amount of data transmitted to earth from the satellite. Then, we would like to address the problematic of online/continuous learning with new data, taking into account issues such as the catastrophic forgetting while minimizing the required energy consumption and the memory footprint.
alain.pegatoquet at univ-cotedazur.fr
benoit.miramond at univ-cotedazur.fr