>allowing valuable AIs to be trained in insecure environments without risking theft of their intelligence
your system involves an unencrypted network and unencrypted data, it would be trivial to train an identical network
the idea of controlling an "intelligence" with a private key is silly. you can achieve effectively the same thing by simply encrypting the weights after training.
Can't someone simply recover the weights of the network by looking at the changes in encrypted loss? I don't think comparisons like "less than" or "greater than" can possibly exist in HE or else pretty much any information one might be curious about can be recovered.
a computer increases performance 10 fold:
what takes 10 minutes now takes 1. we notice a difference of 9 whole minutes! wow!
a new computer increases performance 10 fold:
whoaaaa, that minute of waiting is now merely SECONDS! we saved like almost a whole minute of waiting!
an even newer computer increases performance 10 fold:
nice, thats like noticeabley faster:
today, the newest computer is available:
meh something from like 10 years ago would run this fine...
we are getting to a point where other aspects of performance are also getting harder and harder to notice. things like rendering, supported display sizes and refresh rates, the ability for machines to understand us (voice recognition and computer vision)
I helped create it at a recent hackathon. the ascii is rendered by calculating the intensity of any given pixel, mapping that intensity to a character based on how much that character fills up space ('@' would be bright and '.' would be dark) and then approximating the color of the pixel to fit within the 256 available
I've noticed your implementation uses websockets, which are built on top of TCP. we used UDP and encapsulated everything is a pretty simple to use (and easily the most well documented part of the project) p2plib.c. the fear was that video and audio can get backed up if we were to use TCP. so audio and video were sent via UDP packets and rendered every time a UDP packet was recieved.
well it is 100x40 in number of characters and there are 10 different characters that can take on one of 256 colors. that's 4000 x (8 (256 colors can be represented in 8 bits) + 4 (10 characters in 4 bits)) = 48Kb per frame, with about 20 fps. so maybe 120 KB/s?
the idea of controlling an "intelligence" with a private key is silly. you can achieve effectively the same thing by simply encrypting the weights after training.
Can't someone simply recover the weights of the network by looking at the changes in encrypted loss? I don't think comparisons like "less than" or "greater than" can possibly exist in HE or else pretty much any information one might be curious about can be recovered.