Verification of machine learning models, especially w.r.t. robustness to so-called "adversarial examples" [1], is a major research topic at the moment. There are certainly people worrying about it in the machine learning community, but it appears to be a very difficult problem.
HolyC is great -- default arguments and built-in exception handling are big pluses over C. There are a few shortcomings, though, such as the lack of short-circuiting logical operators and some unfortunate behavior related to variable declarations within nested scopes (just avoid doing that IMO).
I've thought about using this Lisp interpreter to bootstrap a compiler for a new language, probably targeting HolyC. If someone were to write some decent lexer/parser generator tools in HolyC it would go a long way in making something like that possible.
Ideally a HolyC++ would be able to compile normal HolyC without any changes to the source while also adding new stuff and while not being C++ in the process. That all would be very neat.
