Reinforcement learning of motor sequences on motor primitives acquired in the convolutional network-based neocortical model