课程号：

TE196005

课程名：

信息与通讯技术中的机器学习

英文课程名：

Machine Learning in Information and Communication Technology (ICT)

开课单位：

通信工程学院

授课教师：

Heinz Koeppl

责任教师：

肖嵩

助理教师：

肖嵩

学分/学时：

1/16

选课年级：

大二大三大四

选课专业：

计算机、电子信息类

教师简介（中英文）：

Heinz Koeppl 教授出生于奥地利，2001年在格拉茨工业大学获得电机工程博士学位。

2005年获得埃尔温•薛定谔奖学金后，他前往美国加利福尼亚大学伯克利分校获得博士后学位，在担任过洛桑联邦理工学院通讯与计算机科学院非线性系统实验室高级研究员后，受到瑞士苏黎世联邦理工学院的召唤，任职为助理教授。Koeppl教授如今在德国重点理工大学 – 达姆施达特工业大学担任电机工程专业教授，进行授课与科研。主要科研方向为系统与合成生物学的计算科学、信息与通讯技术中的机器学习和基因工程。因为他的科研成就，Koeppl教授曾受邀访问过哈佛医学院，并获得IBM公司的高校科研奖与欧洲科研议会的科研资助。

Since January 2014 Heinz Koeppl is a full professor at the Department of Electrical Engineering and Information Technology at TU Darmstadt with a co-affiliation at the Department of Biology. Prior to that he was a SNSF assistant professor at the Automatic Control Lab of ETH Zurich and a group leader at IBM Research – Zurich.

Heinz went to a school for electrical engineering (HTL) and later got his M.Sc. in physics from Graz Karl-Franzens University (2001) and his Ph.D. in electrical engineering from Graz University of Technology(2004). His PhD thesis was done in collaboration with Infineon Technologies within a Christian Doppler Laboratory. After doing his civil service for the Austrian federal government for a year, he went on as a Schroedinger fellow to UC Berkeley, USA to work with Leon Chua at the EECS department. From fall 2006 to fall 2010 he has been a postdoc at the Laboratory of Nonlinear Systems at the School of Communication and Computer Sciences at EPFL (Ecole Polytechnique Federal de Lausanne) Lausanne, Switzerland. In spring 2008, he did a three month visit at the Department of Systems Biology, Harvard Medical School, Boston USA. He received an IBM Faculty Award (2014) and runs an ERC Consolidator grant since 2018.

课程简介（中英文）：

课程内容

此课程从工程角度介绍机器学习这项领域的形成。介绍重要模型与学习方法，并通过信息技术与通讯技术领域的案例进行配合。知识点包括：

- 概率论与多变量统计基础

- 机器学习问题与模型的分类（监督式、非监督式、生成型、辨别型）

- 复原与分级：理论、方法与信息技术与通讯技术领域的应用

- 维度降低、聚集与大数据分析：通讯与信号处理中的方法与应用

- 概率图像模型：种类、推断与参与估算

- 贝叶斯推断、蒙特卡罗方法与贝叶斯非参数统计的基础

- 凸优化基础：在通讯中的解决方法与应用

- 可扩展型贝叶斯推断的近似算法；在信号处理和信息理论中的应用（比如LDPC编码的解码）

- 隐马尔可夫模型（HMM）：理论、算法与信息技术与通讯技术领域的应用（比如卷积编码的维特比解码）

- 高维统计（大 p 小 n设定），高维数据中的学习依赖结构、学习从观察数据中获得因果关系

- 稀疏估算、随机预测、压缩传感：信号处理领域的理论与应用

- 深度神经网络（深度学习）：模型、学习算法、文库和信息技术与通讯技术领域的应用

学习目标/学习成果

学生可以对信息技术与通讯技术领域中的机器学习的具体工程问题进行阐释与区分。

将此类问题简化为普通机器学习问题并决定适合的解决方法。

从零实现所有所需算法，但也同样熟悉机器学习最先进的文献。

确定所使用的方法的计算复杂性并基于应用限制选择适当的求解算法。

把所需要的方法应用到其它领域，比如生物医学工程的数据分析、社会网络数据分析等。

Content

The module provides an introduction to the emerging field of machine learning from an

engineering perspective. Important models and learning methods are presented and exemplified

through problems from information and communication technology.

•Fundamentals of probability theory and multivariate statistics

•Taxonomy of machine learning problems and models (supervised, unsupervised,

generative, discriminative)

•Regression and classification: theory, methods and ICT applications

•Dimensionality reduction, clustering and big data analytics: methods and application in

communications and signal processing

•Probabilistic graphical models: categories, inference and parameter estimation

•Fundamentals of Bayesian inference, Monte Carlo methods, Bayesian non-parametrics

•Fundamentals of convex optimization: Solution methods and application in

communications

•Approximate algorithms for scalable Bayesian inference; application in signal processing

and information theory (e.g. decoding of LDPC codes)

•Hidden Markov models (HMM): Theory, Algorithms and ICT applications (e.g. Viterbi

decoding of convolutional codes)

•High-dimensional statistics (“large p small n”setting), learning dependency structure in

high-dimensional data, learning causality relations from observational data.

•Sparse estimation, random projections, compressive sensing: Theory and applications in

signal processing

•Deep neural networks (deep learning): Models, learning algorithms, libraries and ICT

applications

Learning objectives / Learning outcomes

Students are able to interpret and categorize specific engineering problems from the ICT domain in terms of machine learning problems.

They are able to reduce such problems to standard machine learning problems and are able to

determine suitable solution methods for them.

They are able to implement all necessary algorithms from scratch, but they are also familiar

with the state-of-the-art libraries in machine learning.

They are able to determine the involved computational complexity of a method and choose an

appropriate solution algorithms based on application constraints.

They are able to apply the acquired methods to other domains, such as data analysis in biomedical engineering, analysis of social network data, etc.

先修课程：

Mathematics（基础工程数学知识）