Research and Analysis of Neural Field Global Architecture and Increase Ability

As a new theory that came into being in 1990’s, information geometry introduces differential geometry method in computer neural science, provides a useful tool for the research of neural networks and information theory, and presents new breakthrough in the research of brain information transformation mechanism Based on information geometry, the global invariant properities of non linear space consisting of neural network models are studied The disintegration ability of the complex neural field architecture is analyzed and proved Finally, a knowledge increasable neural network model is presented The research helps to discover and understand the structure, transformation and locating mechanism of human’s recognition system, stimulate the further neutral network （NN） research, and promote the research level of the connectionist model It also provides important theoretical basis for the understanding of neural network ensembles