Key Words
|
|
|
|
machine learning, neural networks, multilayer perceptrons, gradient descent
|
|
|
|
Course Objectives
|
|
|
|
This course aims to learn mathematical basis of machine learning, such as learning in multilayer perceptrons.
|
|
|
|
Course Goals
|
|
|
|
Focusing on learning in multilayer perceptrons, a typical machine learning system, the historical background and mathematical basis of the theory of machine learning will be explained.
This course provides mathematical basis of machine learning to understand the typical problems in learning in multilayer perceptrons;
1. Vanishing gradient: The gradient in learning becomes locally flat and the learning stagnates.
2. Overfitting: Excess learning specific to the given data reduces the generalization performance.
Recent results on deep learning and on dynamical system theoretic analysis are also discussed.
|
|
|
|
Course Schedule
|
|
|
|
1) Machine learning and neural networks
2) Regression analysis
3) Multilayer perceptron and gradient descent
4) Gradient vanishing and over-fitting
|
|
|
|
Homework
|
|
|
|
To be announced during lecture.
|
|
|
|
Grading System
|
|
|
|
The evaluation will be based on attendance and reports.
Please refer to the class group page at moodle for the details of class implementation.
|
|
|
|
Practical experience and utilization for classes
|
|
|
|
|
|
|
|
|
Condition of tasking the subject
|
|
|
|
|
|
|
|
|
Textbooks
|
|
|
|
|
|
|
|
|
|
Reading List
|
|
|
|
|
|
|
|
|
|
Websites
|
|
|
|
|
|
|
|
|
Website of Laboratory
|
|
|
|
https://www.math.sci.hokudai.ac.jp/~ysato/en/index.html
|
|
|
|
Additional Information
|
|
|
|
The references and textbooks will be announced during the lecture.
|
|
|
|
Update
|
|
|
|
|
|
|
|
|
Class Method
|
|
|
|
|
|
|
|