New Learning Strategies to Neural Networks

Bo Zhang	Ling Zhang
Dept. of Computer Science & Technology	Institute of Artificial Intelligence
Tsinghua University, Beijing	Anhui University, Hefei, Anhui
China	China

Abstract
     Traditional learning approaches in neural networks are based on the bottom-up search technique such as BP. The principle of these approaches is quite simple generally. But they are confronted with the local minima, convergence problem, etc. Therefore, when applying the approaches for solving problems with large-size, the computational complexity is very high. In order to solve the problems, two strategies are used recently. First, instead of the bottom-up search, a top-down constructive learning strategy is used. Secondly, the multi-level learning strategy; i.e., the hierarchical learning, replaces the single-level learning.
     In this talk, we will introduce the principle of one of the constructive learning approaches - covering learning algorithms and hierarchical learning. And the characteristics and applications of the new learning strategies will be discussed.

Biosketches
     Bo Zhang, a computer scientist, is a fellow of Chinese Academy of Sciences and a professor in the Computer Science and Technology Department at Tsinghua University, Beijing, China. In 1958, he graduated from the Automatic Control Department of Tsinghua University. From 1980/02 to 1982/02, he visited University of Illinois at Urbana-Champaign, USA as a scholar. Now he served as the Chairman of the Academic Committee of the Information Science and Technology College at Tsinghua University.
     Zhang Ling, a computer scientist, is a professor of Computer Science Department at Anhui University. In 1961, he graduated from Mathematics and Astronomy Department of Nanjing University. Now he served as the Director of the Artificial Intelligence Institute at Anhui University.
     They both are engaged in the research on artificial intelligence, artificial neural networks, genetic algorithms, fractal, wavelet theory and so on. And they also are engaged in the research applied technology that applies the theories mentioned above into pattern recognition, knowledge engineering, robotics and intelligent control. Their academic achievements won ICL European Artificial Intelligence Prize, the third award of National Natural Science Prize, the first and second award of Science and Technology Progress Prize from the State Educational Commission, the first award of Science & Technology Progress Prize from Electronic Industry Ministry.
     They have jointly published over 60 papers in the fields of artificial intelligence and artificial neural networks, and monographs Theory and Applications of Problem Solving” in Chinese/English version and Theory and Applications of Artificial Neural Networks.