A key thing I do not understand is at 36:00 the binary matrix is supposed to be "smaller" but I do not understand how it is smaller. From a data standpoint, we store bytes, so storing the matrix 3 5 1 4 would take four bytes. We could store a range of numbers from 0-255 using that method. If we wanted to go up to 65k, it would take twice as many bytes. But with storing the binary matrix, if you are storing each 0 and 1 as its own number, that would require 12 bytes. If you are storing it within two bytes, well that means it could only represent a very small range of numbers 0-7, compared to the 256 we could otherwise hold. I do not understand, in any context, how this matrix is supposed to be smaller.
IMPOSSIBLE to listen to for someone with mysophonia. This has to be an A+ grade example of someone who doesn't recognize this affliction. CONSTANT lip smacking. Almost with complete zeal he continues to lipsmack every 1 in 2 sentences. Couldn't stand listening more than 2 minutes before going absolutely insane.
hi, a small question is why it's called "learning with errors"? where does this word learning come from? it's just a set of equations, isn't it? so why "learning", and "learning what"?
Correct me if I'm wrong, say if we falsely assumed y=0, we still find bunch of r's that satisfy the equation, that would still satisfy A.r =cy, making it indistinguishable from y!=0 case
Very informative lecture. Thank you Simons Institute for uploading this lecture on TH-cam.
“If nothing else, just because the dimensions match” @ 29:20 hahaha. Rewatching this video 5 years later and it’s still really good :)
A value or input of 0 at any point should be different from another value in any point in time. Enc (0) != Enc(0)
A key thing I do not understand is at 36:00 the binary matrix is supposed to be "smaller" but I do not understand how it is smaller. From a data standpoint, we store bytes, so storing the matrix 3 5 1 4 would take four bytes. We could store a range of numbers from 0-255 using that method. If we wanted to go up to 65k, it would take twice as many bytes. But with storing the binary matrix, if you are storing each 0 and 1 as its own number, that would require 12 bytes. If you are storing it within two bytes, well that means it could only represent a very small range of numbers 0-7, compared to the 256 we could otherwise hold. I do not understand, in any context, how this matrix is supposed to be smaller.
Smaller in terms of the numeric scale. That is, 1s and 0s are smaller values than 0-8s.
IMPOSSIBLE to listen to for someone with mysophonia. This has to be an A+ grade example of someone who doesn't recognize this affliction.
CONSTANT lip smacking. Almost with complete zeal he continues to lipsmack every 1 in 2 sentences. Couldn't stand listening more than 2 minutes before going absolutely insane.
hi, a small question is why it's called "learning with errors"? where does this word learning come from? it's just a set of equations, isn't it? so why "learning", and "learning what"?
sir can you please provide the link for RLWE based BGV scheme
at 31:30 you say that you can take y = 0 to encrypt 0, but then with the public key and the ciphertext you can find back r and see that y is zero : |
He also mentions that there are much better ways to do it. It is an example that is easy to understand :-)
Correct me if I'm wrong, say if we falsely assumed y=0, we still find bunch of r's that satisfy the equation, that would still satisfy A.r =cy, making it indistinguishable from y!=0 case
44:54
50:27 Homomorphic circuit evaluation
I wonder if you can share your slides.
heat-project.eu/School/Zvika%20Brakerski/HomEncParis.pdf
0:15
disgusting mouth noises but interesting presentation
18:57
24:12
iritating mouth noise, WTF is wrong with him !