It's easy to see that the crucial part of any lattice scheme is the added error. And different schemes seem to use different error distributions, some use Gaussian some use centered Binomial. Though, what do we know currently about the different error distributions used in lattices? I mean which one should be used when, and how does it affect the security of a lattice scheme? What are some important results that specifically study the effect of the error in lattice schemes?
$\begingroup$
$\endgroup$
4
-
$\begingroup$ Could you please give me a reference about some scheme that uses Binomial distribution to sample the noise? :) $\endgroup$– Hilder Vitor Lima PereiraCommented Feb 22, 2018 at 7:26
-
$\begingroup$ @HilderVitorLimaPereira NewHope (eprint.iacr.org/2015/1092.pdf), and Kyber too. $\endgroup$– user4936Commented Feb 22, 2018 at 8:48
-
$\begingroup$ I think you just need something small with enough entropy not to be guessed, I think I saw some schemes using uniform bounded noise also $\endgroup$– Florian BourseCommented Feb 22, 2018 at 9:27
-
$\begingroup$ You should have a quick look at this excellent lattice survey web.eecs.umich.edu/~cpeikert/pubs/lattice-survey.pdf There's some discussion about non-gaussian errors at the end of page 26 $\endgroup$– Florian BourseCommented Feb 22, 2018 at 9:40
Add a comment
|