基於產品汰換速度加快腳步，自眾多試製樣本中抽取出合適的製程參數並依此對現有機台進行調整的思維已逐漸因邁入客製化型態－少量多樣的生產模式而不再適用，如何從樣本數不適用於足夠學習的情境下成為一個不容忽視的小樣本問題。小樣本問題在於樣本數始終都是母體的極小一部分，同時樣本彼此間存在著資料間隙，現有統計方法無法系統性地解決上述問題。整體趨勢擴散技術藉由填補資料間隙並利用母體值域估計技術對資料行為進行預測，並於其中產生虛擬樣本及其對應隸屬函數值以建構適用於小樣本的類神經網路。
然而母體值域估計模型的參數選定仍有著下列問題：偏態係數僅由中心點兩側個數來決定，使不同母體來源的小樣本卻推論出相同母體分配；整體趨勢擴散係數造成過度擴散推論母體分配範圍，讓產生的虛擬樣本可能不符潛在母體特性；原先利用組中點概念對擴散中心點進行定義，於該點附近不一定存在實際樣本等。本研究探討上述缺點並尋求改善方案，加入可能性評估機制重新建構類神經網路，補足原先技術中未能詳加考慮的部分。經兩筆分類案例驗證後發現本研究的確可以在有效縮減擴散範圍前提下有著更好的穩健性，同時亦不因縮減擴散範圍使得準確度下降過多，進一步呈現不同母體來源的樣本在新修正模型參數底下推論出不同母體分配的情形。

This research proposes to improve the parameters in estimating the domain range and membership function of Mega-Trend-Diffusion technique. Beginning, skewness coefficient is only determined by the number of sides in center will make the same estimated population with small sample from different sources. Next, Mega-Trend-Diffusion coefficient results in over-diffusing to real estimated population. Furthermore, there are not necessarily actual samples nearby the diffusing center using class mid-point to define.
To correct above problems one by one with skewness based on distance, coefficient of variation, median and mode. Then joining Plausibility Assessment Mechanism to filter artificial samples producing within the estimated domain range of a small data set. Finally, using artificial samples to train the backpropagation neural network. After verification of two classification cases, this research indeed exists better robustness under reduced diffusing range effectively. Also, showing difference estimated population of samples from different source with improved parameters.

Anthony, M., & Biggs, N. (1997). Computational learning theory: Cambridge University Press.
Cha, S. H. (2007). Comprehensive survey on distance/similarity measures between probability density functions. City, 1(2), 1.
Chongfu, H. (1997). Principle of information diffusion. Fuzzy Sets and Systems, 91(1), 69-90.
Huang, C., & Moraga, C. (2004). A diffusion-neural-network for learning from small samples. International Journal of Approximate Reasoning, 35(2), 137-161.
Jang, J. S. (1993). ANFIS: adaptive-network-based fuzzy inference system. Systems, Man and Cybernetics, IEEE Transactions on, 23(3), 665-685.
Li, D. C., Chang, C. C., & Liu, C. W. (2012a). Using structure-based data transformation method to improve prediction accuracies for small data sets. Decision Support Systems, 52(3), 748-756.
Li, D. C., Chen, C. C., Chang, C. J., & Chen, W. C. (2012b). Employing box-and-whisker plots for learning more knowledge in TFT-LCD pilot runs. International Journal of Production Research, 50(6), 1539-1553.
Li, D. C., Chen, C. C., Chang, C. J., & Lin, W. K. (2012c). A tree-based-trend-diffusion prediction procedure for small sample sets in the early stages of manufacturing systems. Expert Systems with Applications, 39(1), 1575-1581.
Li, D. C., Chen, C. C., Chen, W. C., & Chang, C. J. (2012d). Employing dependent virtual samples to obtain more manufacturing information in pilot runs. International Journal of Production Research, 50(23), 6886-6903.
Li, D. C., Chen, L. S., & Lin, Y. S. (2003). Using Functional Virtual Population as assistance to learn scheduling knowledge in dynamic manufacturing environments. International Journal of Production Research, 41(17), 4011-4024.
Li, D. C., Huang, W. T., Chen, C. C., & Chang, C. J. (2013). Employing virtual samples to build early high-dimensional manufacturing models. International Journal of Production Research, 51(11), 3206-3224.
Li, D. C., & Liu, C. W. (2010). Extending Attribute Information for Small Data Set Classification. IEEE Transactions on Knowledge & Data Engineering(3), 452-464.
Li, D. C., Tsai, T. I., & Shi, S. (2009). A prediction of the dielectric constant of multi-layer ceramic capacitors using the mega-trend-diffusion technique in powder pilot runs: case study. International Journal of Production Research, 47(1), 51-69.
Li, D. C., Wu, C., & Chang, F. M. (2005). Using data-fuzzification technology in small data set learning to improve FMS scheduling accuracy. The International Journal of Advanced Manufacturing Technology, 27(3-4), 321-328.
Li, D. C., Wu, C., & Torng, K. Y. (1997). Using an unsupervized neural network and decision tree as knowledge acquisition tools for FMS scheduling. International Journal of Systems Science, 28(10), 977-985.
Li, D. C., Wu, C. S., Tsai, T. I., & Chang, F. M. (2006). Using mega-fuzzification and data trend estimation in small data set learning for early FMS scheduling knowledge. Computers & Operations Research, 33(6), 1857-1869.
Li, D. C., Wu, C. S., Tsai, T. I., & Lina, Y. S. (2007). Using mega-trend-diffusion and artificial samples in small data set learning for early flexible manufacturing system scheduling knowledge. Computers & Operations Research, 34(4), 966-982.
Tsai, T. I., & Li, D. C. (2008). Approximate modeling for high order non-linear functions using small sample sets. Expert Systems with Applications, 34(1), 564-569.
Tukey, J. W. (1977). Exploratory data analysis: Reading (MA): Addison-Wesley.
Vapnik, V. (2000). The nature of statistical learning theory: Springer, New York.
Wang, Y., & Witten, I. (1997). Inducing model trees for continuous classes. Paper presented at the Proceedings of the Poster Papers of the European Conference on Machine Learning, Prague, Czech Republic.
Zadeh, L. A. (1965). Fuzzy sets. Information and control, 8(3), 338-353.