This was just a quick test to brush up on my PyTorch and test a hypothesis: given a normal and a Poisson distribution, could a very straightforward model identify which distribution a sample came from?
To my surprise, it actually worked surprisingly well with minimal fiddling. This was to prep for an interview that did not move forward, based on a quick and dirty hypothesis that regular traffic might follow one pattern whereas atypical, intrusive traffic might follow another; or in other words, trying to find the atypical signal amongst the noise.
It's not particularly fancy as I threw it together in less than a day, but it might be interesting to follow up on.