A systematic review of federated learning applications for biomedical data
It’s hard enough to prevent algorithmic bias when one can see all the data. It’ll be impossible to detect and mitigate bias when only the weights are shared during the modeling.