Federated Learning: A Way to Get Around Access to Data

The goal of federated learning is to get around this issue that we, that having access to data is sometimes impossible because of privacy concerns and sometimes impossible or undesirable due to engineering constraints. Instead of sending me as a server the data back, I have them send the trained model back, and then, you know, the clue is in the name federated averaging. And now I've got a combined model that in some ways captures all the data that was present on all the phones.