Software And Hardware Design Of No Host Learnable Special-Purpose Neural Networks For Pattern Recognition

Traditional computer has fast and efficient logic operations functions, but its poor image thinking ability and no self-learning ability are not fit for solving pattern recognition problem. Artificial neural networks are nonlinear dynamic systems which are constituted of abundant simple processing unites, their characteristics include parallel computing, memory distributing, self-learning and self-organization, etc. They are used to solve many artificial intelligence problems that are very difficult or impossible to be solved by traditional computer. Combining both features, a no host learnable special-purpose neural networks for pattern recognition has been proposed in this thesis.Based on the high dimensional space covering theory, then we analyze the distribution of samples in high dimensional space. And from the view of geometry, the relationship and arithmetic of points, lines, and...