Title :
Parametric representation of memory surfaces in three-layered neural networks
Author :
Okuda, Toshinobu ; Gouhara, Kazutoshi ; Uchikawa, Yoshiki
Author_Institution :
Dept. of Electron. Mech. Eng., Nagoya Univ., Japan
Abstract :
A "memory surface" of artificial neural networks is defined as a solution set of weights to satisfy a desired input-output pattern. We showed that the memory surface is essential to the supervised learning of the networks. In this paper we show a parametric representation of the memory surface in three-layered neural networks. The explicit expression gives us any point on the memory surface in the weight space.
Keywords :
content-addressable storage; feedforward neural nets; learning (artificial intelligence); input-output pattern; memory surfaces; parametric representation; three-layered neural networks; weight space; Artificial neural networks; Intelligent networks; Multi-layer neural network; Neural networks; Nonlinear equations; Shape; Supervised learning;
Conference_Titel :
Neural Networks, 1993. IJCNN '93-Nagoya. Proceedings of 1993 International Joint Conference on
Print_ISBN :
0-7803-1421-2
DOI :
10.1109/IJCNN.1993.716974
