Re: New lightweight block cipher algorithm



"Jean-Luc Cooke" <jlcooke@xxxxxxxxxx> wrote in message
news:ehj4ar$t2i$1@xxxxxxxxxxxxxxxxxxxxxxxxxxxx
mike3 <mike4ty4@xxxxxxxxx> wrote:
So you don't really think that a cipher can be made from a
genetic algorithm, then (note that I'm not saying that the OP's
cipher is good, I'm just asking a question)?

GA's generally lead to results which are next to impossible to analyse
for weakness. Not to mention you need a success measurment. The
question you ask a GA is "How successful was this combination?" For you
to ask that questions 2^30 times or more, you need a simple way to
answer that.

Find me way to score a possible block cipher (or *any* crypto primitive)
and I'll think about this futher. I haven't found any way.

I will argue that it is possible. If we look at the Wide Trail Strategy we
see that each step of the process is in NP, a GA is generally quite good at
finding a solution for a problem in NP but not in P. Using a GA for each
phase will lead to a good cipher, and since the problem itself is in NP we
can quickly verify the strength of the primitve against the hazards guarded
against by Wide Trail. This does not gaurentee a strong cipher, but it does
mean that a GA can be applied to at least part of the problem.

So I believe there exist GA solutions that assist in many parts of the
design process. One cannot necessarily make an entire cipher this way, but
it can certainly be used to lighten the workload.
Joe


.