Thread

You had me at context specific. Tell me about federated inference?

It is a way to distribute learning of model parameters and inference of wot scores without having to share local user data or preferences. So one can preserve the privacy if needed, but still have reliable estimates. In addition, the client side computation just has to be aware of the subnetwork user cares about, the missing information about the rest of nostr network would come in terms of parameter estimates from other users or relays, which is low dimensional in comparison to raw data.