Web Syllabus

(2018)

Statistical Machine Learning

KOBAYASHI, Yasuyuki

Elective 2 credits

【 Informatics Science〈Correspondence Course〉(Master's Degree Program)･full year】
18-3-1734-1978

１．

Outline

Machine learning has significant performance improvement in recent years, therefore various fields adopt machine learning technique. As machine learning has various methods, therefore, this subject will introduce the concept of statistical machine learning. To consider phenomena with uncertainty, statistical machine learning will give you a stable model estimated from statistical methods with high probability. You will study various basic methods of statistical machine learning with simple exercises with computer.
This subject corresponds to the diploma policy DP3.

２．

Objectives

With the precondition that you have studied artificial intelligence and mathematical statistics for undergraduate, you will be able to understand various methods of statistical machine learning.

３．

Grading Policy

Your overall grade in this class will be decided based on the final examination.
The two reports are required to receive the right to take the final examination, and the results of the two report are not considered for your overall grade.

４．

Textbook and Reference

You do not need any textbooks. The lecture instruction book and materials such as Microsoft Excel macro programs for exercises are posted on LMS.
The important reference documents are as follows:
You can find the other good reference documents, so please obtain them for your own purpose.
1) T. Hastie et al. The Elements of Statistical Learning, 2nd. Ed., (2009), ISBN 978-0387848570.
2) Goto and Kobayashi, Introduction to pattern recognition and machine learning, (2014), ISBN 978-4-339-02479-1 (in Japanese).
3) Sugiyama, Statistical machine learning, (2009), ISBN 978-4-274-50248-4 (in Japanese).
4) Y. Wakui, Bayes statistics as tools, (2009), ISBN 978-4-534-04647-5 (in Japanese).
5) Y. Wakui and S. Wakui, Bayes statistics, (2012), ISBN 978-4-8163-5181-5 (in Japanese).

５．

Requirements (Assignments)

You should study the subject "multivariate analysis" for precondition of statistical machine learning, if you do not understand multivariate analysis at all.
To study statistical machine learning as well as multivariate analysis, it is important for you not only to learn the knowledge and technique, but also to experience the calculation by yourself.
You should review the exercises for each technique of statistical machine learning and apply them to your own problems.

６．

Note

As you will do computer exercises on Microsoft Excel macro programs for Windows, you should prepare a Microsoft Windows PC with a internet connection and confirm whether Microsoft Excel 2007 or later can work on the PC.
This course will be taught in Japanese.

７．

Schedule

１. What is statistical machine learning?

２. Supervised learning (1): linear regression models

３. Supervised learning (2): logistic regression models

４. Supervised learning (2): over-learning phenomena and model selection methods

５. Supervised learning (4): Ridge regression models

６. Supervised learning (5): PLS regression models

７. Supervised learning (6): LASSO regression models

８. Unsupervised learning (1): principal component analysis

９. Unsupervised learning (2): independent component analysis

10. Unsupervised learning (3): cluster analysis and k-means methods

11. Unsupervised learning (4): Gaussian mixture models and EM algorithm

12. Unsupervised learning (5): Graphical modelling and GLASSO

13. Bayesian statistics (1): what is Bayesian statistics?

14. Bayesian statistics (2): Bayes' theorem, etc.

15. Bayesian statistics (3): naive Bayes method, etc.