Emergence of Prediction by Reinforcement Learning Using a Recurrent Neural Network
Figure 2
Localized input signals. Each input signal responds when the object or agent exists around one specific location in or coordinate. This localized representation helps to learn a strong nonlinear mapping.