Abstract
We present a novel growing RBF network structure using SOM in this paper. It consists of SOM and RBF networks respectively. The SOM performs unsupervised learning and also the weight vectors belonging to its output nodes are transmitted to the hidden nodes in the RBF networks as the centers of RBF activation functions, as a result one to one correspondence relationship is realized between the output nodes in SOM and the hidden nodes in RBF networks. The RBF networks perform supervised training using delta rule. Therefore, the current output errors in the RBF networks can be used to determine where to insert a new SOM unit according to the rule. This also makes it possible to make the RBF networks grow until a performance criterion is fulfilled or until a desired network size is obtained. The simulations on the two-spirals benchmark are shown to prove the proposed networks have good performance.
Original language | English |
---|---|
Pages | 107-111 |
Number of pages | 5 |
Publication status | Published - Dec 1 2000 |
Event | 9th IEEE International Workshop on Robot and Human Interactive Communication RO-MAN2000 - Osaka, Japan Duration: Sep 27 2000 → Sep 29 2000 |
Other
Other | 9th IEEE International Workshop on Robot and Human Interactive Communication RO-MAN2000 |
---|---|
Country | Japan |
City | Osaka |
Period | 9/27/00 → 9/29/00 |
Fingerprint
All Science Journal Classification (ASJC) codes
- Hardware and Architecture
- Software
Cite this
Growing RBF structures using self-organizing maps. / Xiong, Qingyu; Hirasawa, Kotaro; Hu, Jinglu; Murata, Junichi.
2000. 107-111 Paper presented at 9th IEEE International Workshop on Robot and Human Interactive Communication RO-MAN2000, Osaka, Japan.Research output: Contribution to conference › Paper
}
TY - CONF
T1 - Growing RBF structures using self-organizing maps
AU - Xiong, Qingyu
AU - Hirasawa, Kotaro
AU - Hu, Jinglu
AU - Murata, Junichi
PY - 2000/12/1
Y1 - 2000/12/1
N2 - We present a novel growing RBF network structure using SOM in this paper. It consists of SOM and RBF networks respectively. The SOM performs unsupervised learning and also the weight vectors belonging to its output nodes are transmitted to the hidden nodes in the RBF networks as the centers of RBF activation functions, as a result one to one correspondence relationship is realized between the output nodes in SOM and the hidden nodes in RBF networks. The RBF networks perform supervised training using delta rule. Therefore, the current output errors in the RBF networks can be used to determine where to insert a new SOM unit according to the rule. This also makes it possible to make the RBF networks grow until a performance criterion is fulfilled or until a desired network size is obtained. The simulations on the two-spirals benchmark are shown to prove the proposed networks have good performance.
AB - We present a novel growing RBF network structure using SOM in this paper. It consists of SOM and RBF networks respectively. The SOM performs unsupervised learning and also the weight vectors belonging to its output nodes are transmitted to the hidden nodes in the RBF networks as the centers of RBF activation functions, as a result one to one correspondence relationship is realized between the output nodes in SOM and the hidden nodes in RBF networks. The RBF networks perform supervised training using delta rule. Therefore, the current output errors in the RBF networks can be used to determine where to insert a new SOM unit according to the rule. This also makes it possible to make the RBF networks grow until a performance criterion is fulfilled or until a desired network size is obtained. The simulations on the two-spirals benchmark are shown to prove the proposed networks have good performance.
UR - http://www.scopus.com/inward/record.url?scp=0034586686&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=0034586686&partnerID=8YFLogxK
M3 - Paper
AN - SCOPUS:0034586686
SP - 107
EP - 111
ER -