3 ms·

I could imagine a scenario where you have many clients sending you information repeatedly, and you’d like to restrict the size of each packet. I think what this

by asxd 5y ago

I could imagine a scenario where you have many clients sending you information repeatedly, and you’d like to restrict the size of each packet. I think what this post is saying is that during the initial client setup you could add an arbitrary amount of shared randomness to increase accuracy while keeping the packet size down. To more directly answer your question, the post mentions federated learning of machine models as an application.

JoachimSchipper 5y ago
Concretely, imagine a sensor node sampling a lot of low-quality sensors frequently. You can afford some setup (to share an AES key, generating as much shared (pseudo-)random as you want), but you want your per-measurement overhead to be small. … it’s not the most practical, and the single-but version seems to be a toy problem, but I can imagine it being useful in some very particular cases?
- asxd 5y ago
  On the hardware side, I think delta-sigma ADCs [1] might solve the same problem. They only sample 1-bit at a time, but they do so at a much faster rate than conventional ADCs. I'm a bit out of my scope here, but I think the concept is really cool. [1] https://en.wikipedia.org/wiki/Delta-sigma_modulation https://en.wikipedia.org/wiki/Delta-sigma_modulation