Title :
Inheriting knowledge in neural networks
Author :
Sayegh, Samir I.
Author_Institution :
Dept. of Phys., Purdue Univ., Fort Wayne, IN, USA
Abstract :
The problem of inheriting knowledge between different networks is examined. Such inheritance allows speeding up training, avoiding some local minima, and coupling fast training networks to fast executing networks. After formulating the general approach, the technique is illustrated and equations are derived for the case of transferring knowledge between a two-layer net and a three-layer net. The equations are written and solved using symbolic algebra techniques for the ease of the XOR
Keywords :
learning (artificial intelligence); neural nets; XOR; fast executing networks; fast training networks; inheriting knowledge; local minima; neural networks; symbolic algebra; three-layer net; two-layer net; Convergence; Density measurement; Equations; Feeds; Forward contracts; Intelligent networks; Joining processes; Least squares approximation; Neural networks;
Conference_Titel :
Neural Networks, 1992. IJCNN., International Joint Conference on
Conference_Location :
Baltimore, MD
Print_ISBN :
0-7803-0559-0
DOI :
10.1109/IJCNN.1992.287082
