I'm sure a Gaussian process could model the equivalent of feedback control; they're pretty flexible (if not as flexible as NNs). The trick is coming up with som

I'm sure a Gaussian process could model the equivalent of feedback control; they're pretty flexible (if not as flexible as NNs). The trick is coming up with something that can learn the feedback controls, or lack thereof, on its own. Unsurprisingly, they're reinventing reinforcement learning.