Background

The field of machine learning constitutes a modern approach to artificial intelligence. It is situated in between computer science, neuroscience, statistics, and robotics, with applications ranging all over science and engineering, medicine, economics, etc. Machine learning algorithms automate the process of learning, thus allowing prediction and decision making machines to improve with experience.

This lecture will cover a contemporary spectrum of supervised learning methods. All lecture material will be in English. In large parts, it is based on this online course, augmented with additional material and a large number of practical exercises.

Topics

The course covers the following topics:

• components of learning: (noisy) data, hypothesis classes, loss function
• working with data: training and test sets, (cross-)validation
• learning theory, VC-bounds, bias-variance tradeoff
• linear methods for classification and regression
• advanced methods like neural networks, support vector machines, ensembles
• regularization, parameter tuning

Participants will learn about the following model classes and learning algorithms:

• linear methods: Perceptron, linear regression, logistic regression, linear hard and soft margin support vector machine, extended with polynomial feature expansion
• k-nearest-neighbor models
• feed forward neural networks, fully connected and convolutional
• non-linear (kernelized) support vector machines
• decision trees, random forests, boosting (ensemble learning)

Organization as an Online Course

The course is organized using the following platforms:

• A Moodle course is used for announcements and for the dissemination of all course material.
• The platform E-Lab is used for the supervision of the exercises during the practical sessions.
• BigBlueButton will be used for interactive video streaming sessions.

Even in "normal" times the course relies on the inverted classroom concept. Students work through the relevant lecture material at home. The material is then consolidated in weekly practical sessions. The course will stick to this format, which should work well for an online course.

Practical course sessions will work as follows: We will start with a video streaming session of about 30-45 minutes to discuss the material of the week. Then we switch to participants working on the exercises, which can be processed in small groups of up to three people. This works fully online from home. The exercises are provided as Jupyter notebooks. Learning groups collaborate through video or voice chat. Questions are asked through E-Lab, and for processing the request the supervisor can enter a group's video conference if necessary.

Course type

Lectures

Credits

6 CP

Term

Summer Term 2020

Lecture

Takes place every week on Thursday from 10:00 to 14:00 in room IA 0/158-79 (PC-Pool 1).
First appointment is on 23.04.2020
Last appointment is on 09.07.2020

The course requires basic mathematical tools from linear algebra, calculus, and probability theory. More advanced mathematical material will be introduced as needed. The practical sessions involve programming exercises in Python. Participants need basic programming experience. They are expected to bring their own devices (laptops).