I’ve recently implemented something similar to this and found the use case and problem very promising. It does seem to be a good step towards privacy & security but given model investigation techniques it probably isn’t as private as it might sound at first glance – particularly on tabular or ill-prepared datasets. Either way, it is an interesting paradigm to adopt and definitely enhances privacy.
You must log in or register to comment.
Here is an interesting blog talking about some of the limitations of federated learning which highlights some of the ‘over promising’ that happens when companies publically speak about the topic: https://www.comparitech.com/blog/information-security/federated-learning/