[choppaface]

Perhaps the idea is they want an APM-like program but for deep learning researchers. The APM program has been pretty successful, and a lot of deep learning progress boils down to tuning architectures and dealing with overfitting and other domain problems. This sounds like a good opportunity for a practitioner, and all the recruiting spin is in place to attract adventurous candidates who might be open to deviating from the traditional grad program grind.