Have them write a program based on Benford's Law. The task combines statistics and programming. The results are interesting, too. Have them find a real dataset online and run it through their program to determine how likely fraud was involved in producing it. I wrote just such a program--it should be tiny--and evaluated population per county in Texas. Was legit! Also have them create their own fraudulent dataset and show that it is fraudulent via their Benford's-Law program.

"FLIF ... is not encumbered by software patents." I wonder what they mean by this. Maybe that the authors have not applied for patents or, as far as they know, FLIF does not infringe on any patents? Regardless, anyone at any time can make a patent claim. Unless someone with big pockets provides patent indemnification, there is no guaranteed that FLIF won't be challenged in the future.

Re facial recognition. I worked in the video-surveillance industry for a few years, and video analytics, especially facial recognition, is a big joke. False positives render it near useless.

Actually, if the design is robust and complete, coding is trivial. For example, I once developed a client for a proprietary FTP server. It took two months to design but just two weeks to code and test. Plus, the design included FSMs that were exploited to develop a test suite for full code coverage. In the end, there were just two coding bugs and one design bug. People are so used to designing on the fly these days while they code. I shake my head in disbelief. Software development has regressed considerably in the last decade or so.

I implemented a floating-point package years ago on a system whose integer CPU did not support floating point. I returned as close to infinity as I could--the largest floating-point number. Mathematically incorrect, but for its purpose, this worked just fine.