簡易檢索 / 詳目顯示
| 研究生： |
陳威昇 Chen, Wei-Sheng |
|---|---|
| 論文名稱： |
類神經網路訓練程序之些許建議 A Few Suggestions for Neural Network Training Algorithm |
| 指導教授： |
胡潛濱
Hwu, Chyanbin |
| 學位類別： |
碩士 Master |
| 系所名稱： |
工學院 - 航空太空工程學系 Department of Aeronautics & Astronautics |
| 論文出版年： | 2003 |
| 畢業學年度： | 91 |
| 語文別： | 中文 |
| 論文頁數： | 111 |
| 中文關鍵詞： | 訓練法則 、類神經網路 |
| 外文關鍵詞： | neural network, training algorithm |
| 相關次數： | 點閱：74 下載：1 |
| 分享至: |
| 查詢本校圖書館目錄 查詢臺灣博碩士論文知識加值系統 勘誤回報 |
倒傳遞網路被廣泛的應用在各個領域，倒傳遞網路通常都採用最陡坡降法來當學習法則。但是，最陡坡降法在使用上有著許多的缺點，例如其網路所需的訓練時間太過沉長，以及容易掉入局部最小值等問題。
本文藉由五種訓練法則(最陡坡降法、共軛梯度法、LM Method、HWO-OWO Method、CGLM Method)去訓練三種類型的例子及一個實際應用例題，從中觀察每種方法的優缺點，並嘗試歸納出一些有用的建議。其中，CGLM Method是將共軛梯度法及LM Method合併使用，希望能比其他方法擁有更佳的收斂性。
Back-propagation neural network is widely applied to every field. It usually adopts the steepest descent method to search minimum of objective function. But the steepest descent method requires lengthy training time and is easy to trap into local minimum.
In order to speed up the convergence, this text uses four different training methods(conjugate gradient method, Levenberg Marquardt method, HWO-OWO method and conjugate gradient-Levenberg Marquardt method) for back-propagation neural network. We observe the shortcomings and advantages of every method and then try to induce some suggestions. Conjugate gradient-Levenberg Marquardt method is to combine the conjugate gradient method and Levenberg Marquardt method. The purpose of this method is to possess better convergence than the other methods.
Charalambous, C.,“Conjugate Gradient Algorithm for Efficient Training of Artificial Neural Networks”, IEEE Proceedings G, Vol. 139, No. 3, pp. 301-310, June 1992.
Chen, H. H. , Manry, M. T. and Chandrasekaran, H.,“A Neural Network Training Algorithm utilizing Multiple sets of Linear Equations”, Neurocomputing, Vol. 25, pp. 55-72, 1999.
Chen, M. S. and Liao, F. H.,“Neural Networks Training Using Genetic Algorithms”, IEEE International Conference, Vol. 3 ,pp. 2436-2441, 1998.
Cho, S. Y. and Chow, T. W. S.,“Training Multilayer Neural Networks using Fast Global Learning Algorithm – Least-Squares and Penalized Optimization Methods”, Neurocomputing, Vol. 25 , pp. 115-131, 1999.
Fun, M. H. and Hagan, M. T.,“Levenberg-Marquardt Training for Modular Networks”, IEEE International Conference, Vol. 1 , pp. 468-473, 1996.
Hagan, M. T. and Menhaj, M. B.,“Training Feedforward Networks with the Marquardt Algorithm”, IEEE Transaction on Neural Networks, Vol.5 , No 6. November 1994.
Haykin, S., Neural Networks A Comprehensive Foundation, Prentice Hall , 1999.
Hwu, C. and Gai, H.S., “Vibration Control of Stiffened Composite Wing Structure”,under preparation.
Liang, Y. C., Feng, D. P., Lee, H. P., Lim, S. P. and Lee, K. H.,“Successive Approximation Training Algorithm for Feedforward Neural Networks”, Neurocomputing, Vol. 42 , pp. 311-322, 2002.
Rao, S. S., Engineering Optimization Theory and PracticeWiley-Interscience, 1996.
Scalero, R. S. and Tepedelenlioglu, N.,“A Fast New Algorithm for Training Feedforward Neural Networks”, IEEE Transaction on Signal Processing, Vol. 40, No. 1, pp.202-210, January 1992.
Towsey, M., Alpsan, D. and Sztriha, L.,“Training a Neural Network With Conjugate Gradient Methods”, IEEE Proc. Neural Networks, Vol. 1, pp. 373-378, 1995.
Yam, Y. F. and Chow, T. W. S.,“Extended backpropagation algorithm”, Electronics Letters, Vol.29, No. 19, pp. 1701-1702, September 1993.
沈宗澤，“以遺傳演算法及共軛梯度法訓練類神經網路”，國立成功大學航空太空工程研究所，2001
施柏屹，“倒傳遞類神經網路學習收斂之初步探討”，國立中央大學機械工程研究所，2000
張吉甫，“類神經網路在寬頻網路管理上之應用”，元智大學資訊管理研究所，2002
葉宜成，“類神經網路模式應用與實作”，儒林，1993
蓋欣聖，“複材機翼結構之振動分析及控制”，國立成功大學航空太空工程研究所，2002
羅華強，“類神經網路-MATLAB的應用”，清蔚科技，2001
校內：立即公開校外：2003-07-14公開