连接主义计算机与神经网络.ppt

Biomimetics Pattern Recognition andMachine Thinking in Image Lab of Artificial Neural Networks Machine Thinking in Image, Institute of Semiconductors, CAS (中科院半导体所神经网络与形象思维实验室) Wang Shoujue(王守觉) 2004.6 1. Development of Information sciences in recent five decades computing speed, storage capacity, quality of intelligence computing speed : thousands billion calculation per second, corresponds to about 1012 times of human brain, as 100 times of total number of human being all over the world. storage capacity : a 100G hard disk corresponds to all information included in a library with 100000 books. quality of intelligence : not even comparable with an animal a、Thinking in Logic b、Thinking in Image the Way to Solve the Image problem 2. Discussion on a Basic Problem of Information Sciences （1）What’s Information In digital world, any information should be described as large amounts of digital numbers a picture, a photo, a speech, a knowledge each of them corresponds to a point in the High Dimensional Space Basic general problem in information sciences ——Point Set Analysis in the High Dimensional Space (2) A brief review of conventional concepts , from point set analysis in the High Dimensional Space A signal in time domain— large amount of digital numbers — a point in the High Dimensional Space “there are no more than n lines existed, which perpendicular to each other, in n-dimensional space” （3）high dimensional geometrical concepts are useful for developing new algorithms for Point Sets Analysis A new method to get sharper picture from a blur picture 3. Biomimetics Pattern Recognition ——application of High Dimensional Geometrical Point Set Analysis in pattern recognition Conventional Pattern Recognition——optimal classification of many classes  Biomimetics Pattern Recognition ——cognizing different classes one by one, by the connectivity of samples in the same class(point set analysis in the High Dimensional Space) （b）human face r