This system very much reminds me of a 'talking machines' episode, where a physics grad student asked a question concerning the definition of "machine learning"

This system very much reminds me of a 'talking machines' episode, where a physics grad student asked a question concerning the definition of "machine learning" when most systems are hand crafted and the actual learning is just statistics of the data put through the system. Here we have a very specific system and they test over different instantiations of the approach, reporting one works better than the other. So what did we learn? For that matter, what did the machine learn? I'll have to read the paper more closely to really get the meat but, it's definitely an interesting problem. On a side note I absolutely love seeing stuff on HN concerning ML being applied to stuff other than "big data"/"deep learning" startups. More links like these please!