1
课程详述
COURSE SPECIFICATION
以下课程信息可能根据实际授课需要或在课程检讨之后产生变动。如对课程有任何疑问,请
联系授课教师。
The course information as follows may be subject to change, either during the session because of unforeseen
circumstances, or following review of the course at the end of the session. Queries about the course should be
directed to the course instructor.
1.
课程名称 Course Title
机器学习系统设计
Machine Learning System Design
2.
授课院系
Originating Department
系统设计与智能制造学院
School of System Design and Intelligent Manufacturing
3.
课程编号
Course Code
SDM374
4.
课程学分 Credit Value
3
5.
课程类别
Course Type
专业选修课
Major Elective Course
6.
授课学期
Semester
秋季 Fall
7.
授课语言
Teaching Language
中英双语
English & Chinese
8.
式(如属团队授课,请列明其
他授课教师)
Instructor(s), Affiliation&
Contact
For team teaching, please list
all instructors
王振坤,助理教授
系统设计与智能制造学院(设计智造学院)
WANG Zhenkun, Assistant Professor
School of System Design and Intelligent ManufacturingSDIM
Email: wangzk3@sustech.edu.cn
9.
实验员/所属学系、联
方式
Tutor/TA(s), Contact
待公布 To be announced
10.
选课人数限额(可不填)
Maximum Enrolment
Optional
待公布 To be announced
2
11.
授课方式
Delivery Method
讲授
Lectures
实验/实习
Lab/Practical
其它(请具体注明)
OtherPlease specify
总学时
Total
学时数
Credit Hours
32
32
0
64
12.
先修课程、其它学习要求
Pre-requisites or Other
Academic Requirements
MA107A 线性代数 A Linear Algebra A
13.
后续课程、其它学习规划
Courses for which this
course is a pre-requisite
NIL
14.
其它要求修读本课程的学系
Cross-listing Dept.
NIL
教学大纲及教学日历 SYLLABUS
本课程旨在提供有关机器学习及数据驱动的设计优化方面的基础知识,从线性回归模型到贝叶斯学习框架。
希望学生掌握基本的机器学习技术,并获得基于模型和数据驱动的系统设计优化技能。
This course aims to provide basic knowledge about machine learning and data-driven design optimization,
from linear regression models to Bayesian learning frameworks. The students are expected to learn basic
machine learning techniques and acquire model-based and data-driven system design optimization skills.
16.
预达学习成果 Learning Outcomes
在学习完成时,学生能够掌握概率分布模型,线性与非线性回归方法,支持向量机,高斯回归模型,人工
经网络,强化学习,贝叶斯优化,代理模型辅助的演化优化模型。
Upon completion of the course, students should master probability distribution models, linear and non-linear
regression methods, support vector machine, Gaussian regression models, artificial neural network,
reinforcement learning, Bayesian optimization, and surrogate model-assisted evolutionary optimization
models.
17.
课程内容及教学日 (如授课语言以英文为主,则课程内容介绍可以用英文;如团队教学或模块教学,教学日历须注明
主讲人)
Course Contents (in Parts/Chapters/Sections/Weeks. Please notify name of instructor for course section(s), if
this is a team teaching or module course.)
第一周:导论
课程介绍
机器学习导论
[实验课] 介绍实验课软件、工具包、网络资源、参考书籍。
第二周:概率分布
二项分布
多项分布
概率与高斯分布
指数家族
[实验课] 学习使用 Matlab 或者 Python 对数据进行分析和预处理
第三周:线性回归模型
线性基函数模型
最大似然估计
线性回归
3
曲线拟合与正则化
预测分布
[实验课] Matlab 或者 Python 实现线性回归任务。
第四周:线性分类模型
最小二乘分类
概率生成模型
逻辑回归与多级逻辑回归
[实验课]利用线性模型,完成数据分类任务。
第五周:项目开题报告
[实验课] 开题报告。
第六周:人工神经网络
前馈网络
反向传播
Jacobian 矩阵与 Hessian 矩阵
正则化
卷积神经网络与对抗生成网络
[实验课] 使用 Pytorch 搭建一个 CNN,完成手写数字识别。
第七周:核方法
最佳分离超平面
分类支持向量机
多核学习
回归支持向量机
[实验课] 实现 SVM,并对数据进行分类。
第八周:混合模型与最大期望学习
K 均值聚类算法
模糊 C 均值聚类
高斯混合模型
最大期望算法
[实验课] 实现聚类算法,完成数据聚类。
第九周:期中考试
[实验课] 期中考试
第十周:隐马尔科夫模型与图方法
隐马尔科夫模型
隐马尔可夫模型的最大期望算法
贝叶斯网络
马尔可夫随机场
图模型中的推断
[实验课]实现最大期望算法
第十一周:马尔可夫决策过程
动态规划
马尔可夫决策过程
部分可观察马尔可夫决策过程
值迭代
策略迭代
[实验课] 使用隐马尔科夫模型,完成对股市的预测。
第十二周:强化学习
Q 学习
时间差分学习
[实验课] 使用 DQN 模型,完成最优路径预测。
4
第十三周:演化算法
基因算法
差分演化算法
粒子群优化算法
基因编程
[实验课] 使用演化算法,完成函数优化。
第十四周:数据驱动的演化优化
代理模型选择
填充准则
多保真模型
在线数据更新
[实验课] 分别使用离线与在线数据驱动的演化优化方法,完成最优设计参数优化。
第十五周:代理模型辅助的多任务与多目标优化
多任务代理模型
多目标代理模型
多目标填充准则
[实验课] 分别使用离线与在线数据驱动的演化优化方法,完成最优设计参数的多任务与多目标优化。
第十六周:总结和复习
[实验课] 复习、答疑。
Week 1: Introduction
Introduction to Course
Introduction to Machine Learning
[Lab] Introduction to the software, tools, online resources that will be used in this module and suggested textbooks.
Week 2: Probability Distributions
Binomial Distributions6
Multinomial Distributions
Gaussian Distributions
Exponential Families
[Lab] Learn to use Matlab or Python to analyze and preprocess data
Week 3: Linear Model for Regression
Linear basis function model
Maximum likelihood estimation
Linear regression
Line fitting and regularization
Predictive Distribution
[Lab] Use Matlab or Python to implement linear regression tasks.
Week 4: Linear Models for Classification
Least Square
Probabilistic generative model
Logistic regression and multi-level logistic regression
[Lab] Use linear models to complete data classification tasks.
Week 5: Proposal Presentations
[Lab] Project proposal.
Week 6: Artificial Neural Network
Feedforward network
Backpropagation
Jacobian matrix and Hessian matrix
Regularization
Convolutional Neural Network and Generative Adversarial Network
[Lab] Use Pytorch to build a CNN to complete handwritten digit recognition.
Week 7: Kernel Method
Optimal separation hyperplane
Support vector machine for classification
Multi-kernel learning
5
Support vector machine for regression
[Lab] Implement SVM and classify data.
Week 8: Mixture Models and Expectation Maximum Learning
K-means clustering algorithm
Fuzzy C-means clustering
Gaussian mixture model
Expectation Maximum Algorithm
[Lab] Implement clustering algorithm and complete data clustering.
Week 9: Mid-term test
[Lab] Mid-term test
Week 10: Hidden Markov Models and Graphical Method
Hidden Markov Model
Expectation Maximum Algorithm of Hidden Markov Model
Bayesian network
Markov Random Field
Inference in the graphical model
[Lab] Implement EM algorithm
Week 11: Markov Decision Process
Dynamic programming
Markov Decision Process
Partially observable Markov decision process
Value Iteration
Policy Iteration
[lab] Use Hidden Markov Model to complete the prediction of the stock market.
Week 12: Reinforcement Learning
Q learning
Time difference learning
[Lab] Use the DQN model to realize the optimal path prediction.
Week 13: Evolutionary Algorithm
Genetic algorithm
Differential evolution algorithm
Particle swarm optimization algorithm
Genetic programming
[Lab] Use evolutionary algorithm to complete function optimization.
Week 14: Data-driven Evolutionary Optimization
Surrogate model selection
Filling criteria
Multi-fidelity model
Online data update
[Lab] Use offline and online data-driven evolutionary optimization to realize the optimization of design parameters.
Week 15: Surrogate-assisted multi-task and multi-objective optimization
Multitask surrogate modeling
Multi-objective surrogate modeling
Multi-objective filling criteria
[Lab] Use offline and online data-driven evolutionary optimization methods to realize multi-task and multi-objective
optimization of optimal design parameters.
Week 16: Summary & Revision
[Lab] Final projects.
18.
教材及其它参考资 Textbook and Supplementary Readings
Textbook:
Bishop C M. Pattern recognition and machine learning[M]. springer, 2006.
6
课程评估 ASSESSMENT
19.
评估形式
Type of
Assessment
评估时间
Time
占考试总成绩百分
% of final
score
违纪处罚
Penalty
备注
Notes
出勤 Attendance
课堂表现
Class
Performance
小测验
Quiz
1-16
Week 1-16
20
NIL
5
Five times
课程项目 Projects
5
Week 5
20
NIL
开题报告
Project proposal
平时作业
Assignments
1-16
Week 1-16
20
NIL
10
Ten times
期中考试
Mid-Term Test
期末考试
Final Exam
期末报告
Final
Presentation
16
Week 16
30
NIL
结题报告
Final project report
其它(可根据需要
改写以上评估方
式)
Others (The
above may be
modified as
necessary)
20.
记分方式 GRADING SYSTEM
A. 十三级等级制 Letter Grading
B. 二级记分制(通过/不通过) Pass/Fail Grading
课程审批 REVIEW AND APPROVAL
21.
本课程设置已经过以下责任人/委员会审议通过
This Course has been approved by the following person or committee of authority