深層樹狀神經網路之初步研究

Primary Study on Deep Tree-like Neural Networks

李俊穎(Chun-Ying Li)

指導教授 ： 謝哲光

共同指導教授 ： 郭英勝(Ying-Sheng Kuo)

義守大學/電機資訊學院/電機工程學系/碩士(2019年)

摘要

決策樹是一種方便且實用之統計分析工具。在資料量較為龐大的情況下，它有能力去挖掘出資料裡可能相對較為複雜的結構，而這些結構可能是傳統統計回歸模型辦不到的。再者，它可以呈現在所有輸入變數中相對重要的變數。基於上述之優點，本論文所要探討的深層樹狀神經網路是模仿相關的決策樹之架構，且其輸入變數是由決策樹所決定。我們將示範如何使用樹狀神經網路於真實資料集之機器學習問題中，並與標準全連結的人工神經網路作比較。

關鍵字

決策樹 ； 神經網路 ； 樹狀神經網路 ； 機器學習

並列摘要

Decision tree is a handy and useful tool for statistical analysis. With relatively large data, it may mine some relatively complicated structure in the data, which may not be revealed using the traditional statistical methods. Moreover, it may uncover relatively important input variables among all input variables. Based on the aforementioned advantages of the decision tree, the proposed deep tree-like neural networks mimic the architecture of the relevant decision trees, and the input variables are determined mostly by decision trees. Some real-world datasets will be used to compare the performances of the proposed tree-like neural networks and the standard densely connected neural networks.

並列關鍵字

decision tree ； neural network ； tree-like neural network ； machine learning

參考文獻

[1] J.G. Hsieh, J.H. Jeng, Y.L. Lin, and Y.S. Kuo, Pathways to Machine Learning and Soft Computing. EHGBooks, USA, 2018.

[2] 謝哲光、鄭志宏、郭英勝、龔志銘、陳軒盈，實用深度學習，滄海，台中，台灣，2018。

[3] J. Brownlee, Deep Learning with Python. (e-book), 2017.

[4] F. Chollet, Deep Learning with Python. First edition. Manning Publications, Shelter Island, New York, USA, 2017.

[5] A. Gulli and S. Pal, Deep Learning with Keras. First edition. Packt Publishing, Birmingham, United Kingdom, 2017.