Federated learning (FL) pools resources of multiple machines that collectively participate in the ML process. While FL parallelizes model training and allows ML to scale better, it introduces the security problem of poisoning when one of the machines inadvertently or deliberately contributes incorrect data to the process. The problem has been extensively studied in the literature, and it continues to be a hot topic of exploration [1].
Recently, the researchers have started to produce tools for visualizing various aspects of the poisoning attack [2]. The goal of this project is to study and improve these tools.
If you are interested, please drop an e-mail to romanvi@ifi.uio.no.
[1] https://ieeexplore.ieee.org/document/9618642
[2] https://ieeexplore.ieee.org/document/10631025