Relations between Sigmoid function′s Polarity and Convergence in B ack Propagetaion Learning
-
- Gotanda Hiromu
- Faculty of Engineering,in Kyushu,Kinki University
-
- Ueda Yoshihiro
- Faculty of Engineering,in Kyushu,Kinki University
-
- Kawasaki Takeshi
- Faculty of Engineering,in Kyushu,Kinki University
-
- Tanaka Nobuyuki
- Faculty of Engineering,in Kyushu,Kinki University
Bibliographic Information
- Other Title
-
- 逆誤差学習におけるシグモイド関数と収束の関係
Search this article
Description
This paper studies how the difference of sigmoidal function′s po larity effects the convergence properties of the error back propagation learning.From simulation results,the convergence properties are summarized as follows.First,in the case of large sized networks,the bipolar sigmoid is superior in both convergence ratios and learning speeds to the unipolar one,and gives good convergence for wider range of initial values than the latter.Then, the convergence properties due to the unipolar sigmoid are apt to depend on the learning rate more than those due to the bipolar one. Finally,in the case of small sized networks,the initial values giving good convergence are smaller for the unipolar sigmoid than for the bipolar one.
Journal
-
- IEICE technical report. Neurocomputing
-
IEICE technical report. Neurocomputing 93 25-32, 1994
The Institute of Electronics, Information and Communication Engineers
- Tweet
Details 詳細情報について
-
- CRID
- 1574231877214207104
-
- NII Article ID
- 110003233281
-
- NII Book ID
- AN10091178
-
- Text Lang
- ja
-
- Data Source
-
- CiNii Articles
