er in the hierarchy corresponding to larger regions). 0 denotes the current observation in the region, C is the state of the higher- layer node, and S and S' denote state variables pertaining to two subsequent time steps. In each node, a statistical learning algorithm is used to predict subsequent states