Self-organizing Multilayer Networks
碩士 === 國立臺灣大學 === 資訊工程學研究所 === 87 === This thesis proposes a new algorithm for training the multilayer networks by the self-organizing approaches which maximize the distances between vectors in each hidden layer. The internal vectors of different prototypes will learn to distribute in the...
| Main Authors: | , |
|---|---|
| Other Authors: | |
| Format: | Others |
| Language: | en_US |
| Published: |
1999
|
| Online Access: | http://ndltd.ncl.edu.tw/handle/21572424731884570723 |
| Summary: | 碩士 === 國立臺灣大學 === 資訊工程學研究所 === 87 === This thesis proposes a new algorithm for training the multilayer networks by the self-organizing approaches which maximize the distances between vectors in each hidden layer. The internal vectors of different prototypes will learn to distribute in the hypercube space and keep away from others as far as possible. The algorithm can avoid the ambiguous binary representation problem which makes back-propagation learning for multilayer perceptrons inefficient. Combining the self-organizing algorithm and back-propagation learning algorithm, the multilayer networks can be trained to give good performance for various problems.
|
|---|