I’ve been thinking about something like this for a while now, glad to see it being worked on.<p>Could federated learning be used as a form of proof-of-work that actually yields valuable data rather than throwaway garbage?<p>Actually, looks like it’s already being done: <a href="https://arxiv.org/abs/1912.11745" rel="nofollow">https://arxiv.org/abs/1912.11745</a>