Hinton points out that ANNs can be trained using reinforcement learning (RL ... with up to 45% memory savings for deep networks.
