一種用於分類與迴歸建模的改良式模糊極限學習機

An Improved Fuzzy Extreme Learning Machine for Classification and Regression Modeling

高子晉(Tzu-Chin Kao)

指導教授 ： 歐陽振森

義守大學/電機資訊學院/資訊工程學系/碩士(2016年)

摘要

本研究針對分類與迴歸建模問題，提出一種改良式模糊極限學習機。此方法主要改良自Wong [2]等人所提出的模糊極限學習機，針對其推廣能力與靈活性易受限制的問題，我們將其常數形態之模糊規則結論推廣為第一階TSK形態，並以妥協式運算元取代其一般化AND運算元。此外，我們亦運用基於遞迴式奇異值分解之最小平方估計值來取代其廣義逆矩陣求解方式，藉以適用於時變系統迴歸建模。為了驗證本研究方法的可行性與優越性，我們與其他方法於分類、時變或非時變系統迴歸建模結果進行比較。實驗結果顯示，本研究所提出改良式模糊極限學習機無論在分類的正確率或迴歸的均方誤差皆有較好且較穩定之表現。

關鍵字

改良式極限學習機 ； 極限學習機 ； 分類 ； 迴歸 ； 妥協式運算元 ； 模糊規則 ； 類神經網路 ； 奇異值分解

並列摘要

In this study, we propose an improved fuzzy extreme learning machine for classification and regression modeling. This approach is mainly an improved version of the fuzzy extreme learning machine proposed by Wong et al. [2]. To alleviate the restricted generalization capability and flexibility problem encountered in Wong’s approach, the original type of rule consequent part, i.e., zero-order Takagi-Sugeno-Kang (TSK) type, is extended to the first-order TSK type, and the original T-norm fuzzy operator, i.e., generalized AND, is replaced by a compensatory fuzzy operator. Besides, the original Moore–Penrose generalized inverse is replaced with a recursive SVD-based least square estimator for solving the problems of regression modeling of time-variant systems. To verify the feasibility and superiority of our approach, we perform several experiments on modeling problems of classification and time-invariant or time-variant systems and make a comparison between our approach and other approaches. Experimental results have shown our approach produces the higher classification accuracy for classification problems and the lower mean squared errors for regression problems, and possesses the better stability.

並列關鍵字

Improved fuzzy extreme learning machin ； Extreme learning machine ； Classification ； regression ； Compensatory operator ； Fuzzy rule ； Neural network ； Singular value decomposition

參考文獻

[1] Guang-Bin Huang, Qin-Yu Zhu, and Chee-Kheong Siew, “Extreme Learning Machine: A New Learning Scheme of Feedforward Neural Networks.” IEEE International Joint Conference, vol. 2, pp. 985-990, 2004.

[2] Shen Yuong Wong, Keem Siah Yap, Hwa Jen Yap, Shing Chiang Tan, and Siow Wee Chang, “On Equivalence of FIS and ELM for Interpretable Rule-Based Knowledge Representation.” IEEE Transactions on Neural Networks ond Learning Systems, vol. 26, no. 7, pp. 1417-1430, 2015.

[3] Wang, Y. G., Cao, F. L., & Yuan, Y. B. “A study on effectiveness of extreme learning machine.” Neurocomputing, vol. 74, no. 16, pp. 2483–2490, 2011.

[4] Chen, Z. X. X., Zhu, H. Y. Y., & Wang, Y. G. G. “A modified extreme learning machine with sigmoidal activation functions.” Neural Computing & Applications, vol. 22, no. 3, pp. 541–550, 2013.

[5] Huang, G.-B., Chen, L., & Siew, C.-K. “Universal approximation using incremental constructive feedforward networks with random hidden nodes.” IEEE Transactions on Neural Networks, vol. 17, no. 4, pp. 879–892. 2006.