臺灣博碩士論文加值系統

在過去幾年中，機器學習在各種應用中的成功，導致機器學習系統的需求快速增長。典型的機器學習模型包括以下過程：從攝取數據到預處理、優化、預測結果，每個步驟都由人工控制和執行。機器學習現在的成功必須要靠有經驗的專家選擇出合適的特徵、模型、優化、評價有關的重要步驟。然而要讓機器學習未來能夠大量的被運用需要找到一個通用且易用的方法才行，這種將機器學習自動化的研究領域被稱之為AutoML全稱是Automated Machine Learning，是2014年以來機器學習和深度學習領域最炙手可熱的領域之一。自動化機器學習之議題日趨重要，Google AI應用AutoML來尋找更好的捲積網絡替代架構，對自動駕駛汽車中的圖像進行分類。營銷部門的報告速度提高了30％，準確度提高了8％，令人印象深刻。
目前市場上有很多工具與平台聲稱具有AutoML，但大多只提供超參數調校，卻沒有資料清整、資料探索或是資料轉置的功能，有些著重在Atuo ML的功能、部分則是強調在使用者體驗(UX)或是提供更多資料視覺化的服務。不約而同的是，他們都把智慧自動化模型當作最重要的任務。本論文旨在了解autoML概念及應用當前框架，透過mnist手寫辨識圖像，了解其中運作方法及效能。

In the past few years, the success of machine learning in a variety of applications has led to a rapid increase in the demand for machine learning systems. A typical machine learning model includes the following processes: from ingesting data to pre-processing, optimization, and predicting results, each step is manually controlled and executed. The current success of machine learning must rely on experienced experts to select the appropriate features, models, optimization, and evaluation of important steps. However, in order for machine learning to be used in large numbers in the future, it is necessary to find a universal and easy-to-use method. The research field that automates machine learning is called AutoML. The full name is Automated Machine Learning, which is machine learning and since 2014. One of the hottest areas of deep learning. The topic of automated machine learning is becoming more and more important. Google AI uses AutoML to find a better alternative architecture for convolutional networks to classify images in autonomous vehicles. The marketing department’s reporting rate has increased by 30% and accuracy has increased by 8%, which is impressive.
At present, there are many tools and platforms on the market that claim to have AutoML, but most of them only provide hyper-parameter tuning, but there is no data clearing, data exploration or data transposition function. Some emphasize the function of autoML, and some emphasize it. User experience (UX) or a service that provides more information visualization. Invariably, they all regard the intelligent automation model as the most important task. This thesis aims to understand the current framework of autoML concepts and applications, and to identify images and understand the operation methods and performance through mnist handwriting.

中文摘要 III
Abstract IV
Acknowledgement V
Table of Contents VI
List of Figures VIII
List of Tables IX
1. Chapter 1 Introduction 1
1.1 Background and Motivation 1
1.2 Objective 2
1.3 Organization of the Thesis 3
2. Chapter 2 Literature Review 4
2.1 Machine Learning status 4
2.2 Automated Machine Learning (AutoML) 5
2.2.1 The definition of AutoML 5
2.2.2 Work of AutoML 6
2.2.3 Framework of AutoML (at the present stage) 9
3. Chapter 3 Methodology Design 12
3.1 Experimental operation process 12
3.2 Description of Data 13
3.3 Google Cloud automl 15
3.4 Autokeras 20
3.5 Manual modeling 26
4. Chapter 4 Implementation and Analysis 31
4.1 Google cloud autoML 31
4.2 Autokeras 34
4.3 Manual modeling 35
4.4 Comprehensive discussion 37
Manual modeling 37
5. Chapter 5 Conclusion and Future Research 38
References 39

[1] C. Metz. "Tech Giants Are Paying Huge Salaries for Scarce A.I. Talent." (accessed.
[2] C. Boyd. "AI scientists: How can companies deal with the shortage of talent?" (accessed.
[3] J. Kahn. "Just How Shallow is the Artificial Intelligence Talent Pool?" (accessed.
[4] W. Team. "AutoML: Automating the design of machine learning models for autonomous driving." (accessed.
[5] M. Shih. "Automated Machine Learning: An Agile Approach to AI Deployment." (accessed.
[6] T. Mitchell, "Machine Learning," 1997.
[7] Y. Quanming et al., "Taking human out of learning applications: A survey on automated machine learning," arXiv preprint arXiv:1810.13306, 2018.
[8] 武耀文. "AI平台-AutoML-2019年几个靠谱的AutoML框架." (accessed.
[9] F. Baradel, C. Wolf, J. Mille, and G. W. Taylor, "Glimpse clouds: Human activity recognition from unstructured feature points," in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2018, pp. 469-478.
[10] J. Friedman, T. Hastie, and R. Tibshirani, The elements of statistical learning (no. 10). Springer series in statistics New York, 2001.
[11] R. A. Fisher, "The use of multiple measurements in taxonomic problems," Annals of eugenics, vol. 7, no. 2, pp. 179-188, 1936.
[12] A. B. Alexander Allen "Automatic Machine Learning (AutoML) Landscape Survey," ed, 2018.
[13] A. Kaul, S. Maheshwary, and V. Pudi, "Autolearn—Automated feature generation and selection," in 2017 IEEE International Conference on Data Mining (ICDM), 2017: IEEE, pp. 217-226.
[14] M. Elad and M. Aharon, "Image denoising via sparse and redundant representations over learned dictionaries," IEEE Transactions on Image processing, vol. 15, no. 12, pp. 3736-3745, 2006.
[15] "AutoML: Automatic Machine Learning." http://docs.h2o.ai/h2o/latest-stable/h2o-docs/automl.html (accessed.
[16] L. Bottou and O. Bousquet, "The tradeoffs of large scale learning," in Advances in neural information processing systems, 2008, pp. 161-168.
[17] Q. V. Le, J. Ngiam, A. Coates, A. Lahiri, B. Prochnow, and A. Y. Ng, "On optimization methods for deep learning," in Proceedings of the 28th International Conference on International Conference on Machine Learning, 2011: Omnipress, pp. 265-272.
[18] B. Zoph and Q. V. Le, "Neural architecture search with reinforcement learning," arXiv preprint arXiv:1611.01578, 2016.
[19] B. Zoph, V. Vasudevan, J. Shlens, and Q. V. Le, "Learning transferable architectures for scalable image recognition," in Proceedings of the IEEE conference on computer vision and pattern recognition, 2018, pp. 8697-8710.
[20] E. Real, A. Aggarwal, Y. Huang, and Q. V. Le, "Regularized Evolution for Image Classifier Architecture Search," arXiv e-prints. [Online]. Available: https://ui.adsabs.harvard.edu/abs/2018arXiv180201548R
[21] "auto-sklearn." https://automl.github.io/auto-sklearn/master/ (accessed.
[22] "TPOT." https://github.com/EpistasisLab/tpot (accessed.
[23] "Auto-Keras." https://autokeras.com/ (accessed.
[24] "Cloud AutoML." https://cloud.google.com/automl/?hl=Zh-TW (accessed.
[25] A. Balaji and A. Allen, "Benchmarking Automatic Machine Learning Frameworks," arXiv preprint arXiv:1808.06492, 2018.
[26] "Tensorflow." https://www.tensorflow.org/tutorials (accessed.
[27] "THE MNIST DATABASE." http://yann.lecun.com/exdb/mnist/ (accessed.
[28] ariesjzj. "單機玩轉神經網絡架構搜索(NAS) - Auto-Keras學習筆記." https://jinzhuojun.blog.csdn.net/article/details/88934040 (accessed.