Abstract
Learning algorithms are described for layered feedforward type neural networks, in which a unit generates a real-valued output through a logistic function. The problem of adjusting the weights of internal hidden units can be regarded as a problem of estimating (or identifying) constant parametes with a non-linear observation equation. The present algorithm based on (he extended Kalman filter has just the time-varying learning rate, while the well-known back-propagation (or generalized delta rule) algorithm based on gradient descent has a constant learning rate. From some simulation examples it is shown that when a sufficiently trained network is desired, the learning speed of the proposed algorithm is faster than that of the traditional back-propagation algorithm.
| Original language | English |
|---|---|
| Pages (from-to) | 753-768 |
| Number of pages | 16 |
| Journal | International Journal of Systems Science |
| Volume | 22 |
| Issue number | 4 |
| DOIs | |
| Publication status | Published - 1991 |
| Externally published | Yes |
Fingerprint
ASJC Scopus subject areas
- Control and Systems Engineering
- Computer Science Applications
- Theoretical Computer Science
- Computational Theory and Mathematics
- Management Science and Operations Research
Cite this
Learning algorithms of layered neural networks via extended kalman filters. / Watanabe, Keigo; Fukuda, Toshio; Tzafestas, Spyros G.
In: International Journal of Systems Science, Vol. 22, No. 4, 1991, p. 753-768.Research output: Contribution to journal › Article
}
TY - JOUR
T1 - Learning algorithms of layered neural networks via extended kalman filters
AU - Watanabe, Keigo
AU - Fukuda, Toshio
AU - Tzafestas, Spyros G.
PY - 1991
Y1 - 1991
N2 - Learning algorithms are described for layered feedforward type neural networks, in which a unit generates a real-valued output through a logistic function. The problem of adjusting the weights of internal hidden units can be regarded as a problem of estimating (or identifying) constant parametes with a non-linear observation equation. The present algorithm based on (he extended Kalman filter has just the time-varying learning rate, while the well-known back-propagation (or generalized delta rule) algorithm based on gradient descent has a constant learning rate. From some simulation examples it is shown that when a sufficiently trained network is desired, the learning speed of the proposed algorithm is faster than that of the traditional back-propagation algorithm.
AB - Learning algorithms are described for layered feedforward type neural networks, in which a unit generates a real-valued output through a logistic function. The problem of adjusting the weights of internal hidden units can be regarded as a problem of estimating (or identifying) constant parametes with a non-linear observation equation. The present algorithm based on (he extended Kalman filter has just the time-varying learning rate, while the well-known back-propagation (or generalized delta rule) algorithm based on gradient descent has a constant learning rate. From some simulation examples it is shown that when a sufficiently trained network is desired, the learning speed of the proposed algorithm is faster than that of the traditional back-propagation algorithm.
UR - http://www.scopus.com/inward/record.url?scp=0026136149&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=0026136149&partnerID=8YFLogxK
U2 - 10.1080/00207729108910654
DO - 10.1080/00207729108910654
M3 - Article
AN - SCOPUS:0026136149
VL - 22
SP - 753
EP - 768
JO - International Journal of Systems Science
JF - International Journal of Systems Science
SN - 0020-7721
IS - 4
ER -