[Cryptography] "Practical Construction and Analysis of Pseudo-Randomness Primitives"

[Caleb Cannon]

Cool find. I was looking for just this some years ago for a small project
that also might be of interest here. At the time, I was dissatisfied with
the performance of PRNGs used in shader based procedural graphics, so I
wrote a PRNGG. That is, a pseudo random number generator generator. The
software is basically a genetic algorithm that operates on a parse tree
representing a PRN generating function. It could change constants, insert
nodes, change nodes from one operator to another, etc. Fitness tests were
based on performance, memory use, and random number quality. It worked, but
the part that I had the most trouble with was testing the quality of the
random number generation. I haven't touched the project in years, but I
have wondered if a utility like that has some use beyond my original intent.

-Caleb

P.S. The software was also interesting for a more abstract reason. Because
the GA uses random numbers, and the used PRNG can be improved by the
program, the program can recompile itself using the improved random number
generator and thus become a genuinely self-improving piece of software. I'm
sure there has been some talk about this sort of thing in the GA community,
but my research (which lacks great skill) didn't turn anything up.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://www.metzdowd.com/pipermail/cryptography/attachments/20210621/3d5719a2/attachment.htm>