在小樣本的領域中，自從整體趨勢擴散( Mega-Trend-Diffusion, MTD )方法被提出後，已經在多項研究中顯示其有效性和實用性。近期隨著深度學習議題熱絡，有學者結合生成對抗網路與MTD方法，發表新興的WGAN_MTD數據生成架構，此方法透過MTD值域推估限制生成對抗網路的樣本值域，生成出有效的虛擬樣本。但含有離群值的真實樣本產生出的虛擬樣本有效性具有爭議，且Wasserstein GAN ( Wasser-stein Generative Adversarial Network, WGAN )的修剪權重方法已被證實會影響模型訓練的穩定性。
本研究利用合鬚圖及限縮MTD的懲罰項方法，降低WGAN_MTD受離群值的影響，並以卷基層取代神經元捕捉局部資訊特徵，減少輸入輸出層之間的參數降低模型複雜度，而生成對抗網路模型則引用WGAN with Gradient Penalty ( WGAN-GP )取代WGAN，提高訓練的穩定度、精確度。實驗結果證明，在含有離群值的小樣本數據中，合鬚圖及限縮MTD的懲罰項方法都能提升生產模型的準確度。
經改良後的研究模型在少變量特徵的條件下，對比未生成虛擬樣本與使用WGAN_MTD生成虛擬樣本，本研究模型都能提供更好且更準確的虛擬樣本，解決小樣本資料不足的問題。

In the field of small-sample domains, since the introduction of the Mega-Trend Dif-fusion (MTD) method, its effectiveness and practicality have been demonstrated in vari-ous studies. Recently, with the popularity of deep learning, scholars have combined gen-erative adversarial networks (GAN) with the MTD method and proposed a novel frame-work called WGAN_MTD for data generation. This method utilizes the MTD value ran-ge estimation to restrict the sample value range of the generative adversarial network, thereby generating effective synthetic samples. However, the validity of generating virtual samples using real samples containing outliers remains controversial, and the weight clip-ping method in WGAN has been shown to affect the stability of model training.
In this study, we propose the use of boxplot and a penalization term to limit the in-fluence of outliers on WGAN_MTD. We also replace the neurons with convolutional lay-ers to capture local information features and reduce the complexity of the model by reduc-ing the number of parameters between the input and output layers. Additionally, we adopt the WGANGP method instead of WGAN to improve training stability and precision in the generative adversarial network model. Experimental results demonstrate that both the mustache diagrams and the penalization term for limiting MTD enhance the accuracy of the generated models in small-sample data containing outliers.

Arjovsky, M., Chintala, S. & Bottou, L. (2017). Wasserstein gan. arXiv preprint arXiv:1701.07875.
Chawla, N. V., Bowyer, K.W., Hall, L.O., & Kegelmeyer, W.P. (2002). SMOTE: SYn-thetic minority over-sampling technique. Journal of Artificial Intelligence Research, 16, 321-357.
Douzas, G., Lechleitner, M. & Bacao, F. (2022). Improving the quality of predictive mod-els in small data GSDOT: A new algorithm for generating synthetic data. PLoS ONE, 17(4), e0265626.
Efron, B. & Tibshirani, R. (1994). An Introduction to the Bootstrap. CRC Press, 1994.
Gong, H. F., Chen, Z. S., Zhu, Q. X. & He, Y. L. (2017). A Monte Carlo and PSO based virtual sample generation method for enhancing the energy prediction and energy op-timization on small data problem: An empirical study of petrochemical industries. Ap-plied Energy, 197, 405-415.
Goodfellow, I., Pouget-Abadie, J., Mirza, M., Xu, B., Warde-Farley, D., Ozair, S., Cour-ville, A. & Bengio, Y. (2014). Generative adversarial nets. Advances in Neural In-formation Processing Systems, 27, 2672-2680.
Gulrajani, I., Ahmed, F., Arjovsky, M., Dumoulin, V. & Courville, A. (2017). Improved training of wasserstein GANs. Advances in Neural Information Processing Systems, 30, 5768-5778.
Huang, C. & Moraga, C. (2004). A diffusion-neural-network for learning from small samples. International Journal of Approximate Reasoning, 35(2), 137-161.
Li, D. C., Chen, S. C., Lin, Y. S. & Huang, K. C. (2021). A Generative Adversarial Net-work Structure for Learning with Small Numerical Data Sets. Applied Science-Basel, 11(22), 10823.
Li, D. C., Chen, C. C., Chang, C. J., & Chen, W. C. (2012). Employing box-and-whiskerplots for learning more knowledge in TFT-LCD pilot runs. Interna-tional Journal of Production Research, 50(6), 1539-1553.
Li, D. C., Wu, C. S., Tsai, T. I., & Lina, Y. S. (2007). Using mega-trend-diffusion and ar-tificial samples in small data set learning for early flexible manufacturing system scheduling knowledge. Computers & Operations Research, 34, 966-982.
Li, S., Li, J. Z., He, H., Ward, P., & Davies, B. J. (2011). WebDigital: A Web-based hybrid intelligent knowledge automation system for developing digital marketing strategies. Expert Systems with Applications, 38(8), 10606-10613.
Lin, Y. S. & Li, D. C. (2010). The Generalized-Trend-Diffusion modeling algorithm for small data sets in the early stages of manufacturing systems. European Journal of Operational Research, 207(1), 121-130.
Lin, C. F. & Fu, C. S. (2018). Evaluating online advertising effect: An approach integrat-ing means-end conceptualization and similarity analysis. Electronic Commerce Re-search and Applications, 32, 1-12.
Shen, L. & Qian, Q. (2022). A virtual sample generation algorithm supporting machine learning with a small-sample dataset: A case study for rubber materials. Computa-tional Materials Science, 211, 111475.
Sun, B., Wu, Z. Y., Feng, Q., Wang, Z. L. Ren, Y., Yang, D. Z. & Xia, Q. (2023). Small Sample Reliability Assessment with Online Time-Series Data Based on a Worm Wasserstein Generative Adversarial Network Learning Method. IEEE Transactions on Industrial Informatics, 19(2), 1207-1216.
Tukey, J. W. (1977). Exploratory data analysis (Vol. 2). Reading, Mass.
Wang, Y. F., Dong, X. S., Wang, L. X., Chen, W. D. & Zhang, X. J. (2022). Optimizing Small-Sample Disk Fault Detection Based on LSTM-GAN Model. ACM Transac-tions on Architecture and Code Optimization, 19(1), 1-24.
Yoon, J., Jarrett, D. & Schaar, M. (2019). Time-series Generative Adversarial Networks. 33rd Conference on Neural Information Processing Systems (NeurIPS 2019), Van-couver, Canada.
Yu, X., He, Y., Xu, Y., & Zhu, Q. (2019). A Mega-Trend-Diffusion and Monte Carlo based virtual sample generation method for small sample size problem. Journal of Physics: Conference Series, 1325, 012079.
Zekan, M., Tomicic, I. & Schatten, M. (2022). Low-sample classification in NIDS using the EC-GAN method. Journal of Universal Computer Science, 28(12), 1330-1346.
Zhong, X. & Ban, H. (2022). Pre-trained network-based transfer learning: A small-sample machine learning approach to nuclear power plant classification problem. Annals of Nuclear Energy, 175, 109201.
Zhu, B., Chen, Z. S. & Yu, L. A. (2016). A novel mega-trend-diffusion for small sample. CIESC Journal, 67(3), 820-826.