An explanation of reasoning neural networks

Reasoning Neural Networks (RN) adopts the layered feedforward network structure, and its learning algorithm belongs to the weight-and-structure-change category of learning algorithm. In this paper, we firstly explain that, in the layered feedforward network, the essential characteristic of the mapping between two consecutive layers is the level-adjacent mapping, in which level-adjacent patterns in the previous-layer space are mapped to similar patterns in the latter-layer space. Then, we explain how RNʹs learning algorithm handles the undesired predicaments associated with the back propagation learning algorithm.