| 课程号 |
04833610 |
学分 |
2 |
| 英文名称 |
Machine Learning for Time Series Analysis – Statistical Models and Deep Learning |
| 先修课程 |
概率与统计，线性代数，优化 |
| 中文简介 |
本课程介绍概率模型和深度学习等机器学习模型。本课程欢迎对基于机器学习/概率模型的时序分析和预测感兴趣同学参加，欢迎计算机科学、统计学、经济学、财政学、电子工程、生物、物理等不同学科背景的同学参加。本课程没有编程技巧的要求，但是编程有益于动手做实验。本课程涵盖了几个比较普及的时序模型，如向量回归模型、ARIMA模型、隐马尔科夫模型、Karman滤波等模型，还包括比较先进的模型如神经网络（长短时记忆网络、递归神经网络、门递归神经网络）、支持向量回归模型、Hawkes过程、稀疏VAR模型等。在本课程的最后，我们希望学生能够做到：（1）理解时序模型的数学公式。（2）将时序模型应用到真实的数据中。（3）具有能够为时序应用开发出新颖的机器学习模型的能力，并以此发表论文。 |
| 英文简介 |
The course aims to introduce machine learning models, including both statistical models and deep learning, to students in various science disciplines such as computer science, statistics, economics, finance, electronic engineering, biology, physics etc., who are interested in machine learning and statistical models for time series analysis and forecasting. Enrolled students should have basic knowledge in statistics and probability, linear algebra, optimization. No programming skills are required, but could be helpful for hands-on exercise. The class covers popular time series models, including vector-regressive models (VAR), ARIMA models, hidden Markov models, Karman filtering, as well as advanced models, such as neural network models (Long Short-term memory neural networks, recurrent neural networks, gated recurrent neural networks), support vector machine regression, Hawkes processes, sparse VAR models etc. At the end of the course, the students are expected to be able to do the following: (1) understanding the mathematical formulation of time series models; (2) apply time series models to real-application data; (3) potential of developing novel machine learning models for time series applications for publications. |
| 开课院系 |
信息科学技术学院 |
| 通选课领域 |
|
| 是否属于艺术与美育 |
否 |
| 平台课性质 |
|
| 平台课类型 |
|
| 授课语言 |
中文 |
| 教材 |
Time Series Analysis,James D. Hamilton,Princeton Press,1994；
Deep Learning,Ian Goodfellow and Yoshua Bengio and Aaron Courville,MIT Press,2016；
|
| 参考书 |
|
| 教学大纲 |
无
Session 1: Introduction to Time Series
【Description of the Session】(purpose, requirements, class and presentations scheduling, etc.)
Definition of time series, stationary and non-stationary time series, white noise
Applications of time series analysis and forecasting
Introduction of basic time series models, moving average, auto-regression, ARMA models and extensions
【Questions】
What is time series? What is stationary time series and non-stationary time series? What is the basic models for time series analysis
【Readings, Websites or Video Clips】
1. Hamilton Ch 1-4, 11
【Assignments for this session (if any)】
Exercise 1. Questions on basic definitions of time series and stationary time series
Exercise 2. Hands-on exercise on running R-code for moving average, auto-regression, ARMA models
Session 2: State-space models Date: 7/24/2018
【Description of the Session】(purpose, requirements, class and presentations scheduling, etc.)
Introduction of state-space models, including hidden Markov model and Kalman filter
【Questions】
What is state-space models? What is hidden Markov model and Kalma filter? When should we apply these models
【Readings, Websites or Video Clips】
1. Hamilton Ch 5
2. Reading- Rabiner 1986
【Assignments for this session (if any)】
Exercise 3. Questions on hidden Markov models and Kalman filtering
Exercise 4. Hands-on exercise on hidden Markov models applying to text modeling
Session 3: Neural Network models for time series Date: 7/25/2018
【Description of the Session】(purpose, requirements, class and presentations scheduling, etc.)
Introduction of basic neural network models, recurrent neural networks (RNN), long short-term memory neural networks, gated recurrent neural networks
【Questions】
What is RNN? What type of properties does RNN capture in time series? How to apply RNN for time series forecasting and prediction
【Readings, Websites or Video Clips】
1. Goodfellow et al, 2016 Chapter 4-6, 10
【Assignments for this session (if any)】
Exercise 5: Hands-on exercise on RNN for time series analysis
Session 4: Sparse VAR models and Granger causality Date: 7/26/2018
【Description of the Session】(purpose, requirements, class and presentations scheduling, etc.)
Introduction of lasso, VAR models, sparse VAR models, Granger causality and extensions
【Questions】
How can we learn temporal dependencies? How can we address problems in practical applications, such as nonstationary, irregular time series, relational time series?
【Readings, Websites or Video Clips】
1. Reading- Liu 2018
【Assignments for this session (if any)】
Exercise 6: Hands-on exercise on sparse-VAR models for time series dependence analysis
Session 5: Hawkes Process, and Support Vector Regression Date: 7/27/2018
【Description of the Session】(purpose, requirements, class and presentations scheduling, etc.)
Introduction of Poisson process, Hawkes processes, support vector machines, and support vector regression
【Questions】
How can we model stochastic data? How can we go beyond linear predictors?
【Readings, Websites or Video Clips】
1. Reading- Laub et al, 2015
2. Reading- Burges, 1998
【Assignments for this session (if any)】
Exercise 7. Hands-on exercise on Hawkes process models
文献阅读，课堂讨论，在线讨论和课后练习
考勤和课堂讨论：25%，书面作业：45%，考试：30%
|
| 教学评估 |
王亦洲：
学年度学期：17-18-3，课程班：机器学习与时间序列分析1，课程推荐得分：4.21，教师推荐得分：4.01，课程得分分数段：80-85；
|
|