In this course of seminar lectures, we build upon previous knowledge of machine learning to dwell on more advanced topics.

Students are assumed to have a background in Machine Learning. The easiest way to achieve this is to have followed our own Introduction to Machine Learning course, but having followed lectures at other institutes or in MOOC is also considered a valid substitute (although verification may be required).

The seminars will be held by the students, for the students. The students will be divided into groups of 2, each of which will have to give an actual 90 minute lecture on a specific topic. The goal is to have each group spend time researching into his own topic and design an interactive lecture (with slides) with which to explain the topic to the fellow students. The lecture is required to pick up from the general machine learning knowledge which the students are assumed to have, and then gradually explain the topic at hand and going into a technical depth appropriate for a real lecture in machine learning.

Lectures should contain interactive components such as examples and open questions posed to and from the audience. About 60 minutes should be devoted to actual contents, and 30 minutes to exercises and questions and other interactive components. It is important that the division of each lecture between the two students giving the lecture should be dynamic; it won’t be acceptable to have one student explain the first half and the other the second half. Both students should be prepared on the whole lecture and be able to pick up from any point, often switching the role of the speaker.

After all the lectures have been given, the students will be given a simple test involving all of the topics which were explained during the course of the seminars. This test will not be used to grade the students, but is rather an incentive for the students to give good lectures and to focus on the other lectures as well.

Students should finally submit a final report on their specific topic, which can be thought of as a written version of the tutorial for their own topic. There are no hard constraints on the length of such documents; a general guideline is that anything less than 5 pages or more than 10 pages will not be considered appropriate. Grading will be based on the quality of the lecture and the report produced by each student pair.

The available topics for this year are the following:

- Spectral Learning (also in HMMs)
- Active Learning and Experimental Design
- Non-parametrics
- Variational Inference
- Deep Learning
- Markov-Chain Monte-Carlo (also Monte-Carlo Tree Search)
- Chinese-Restaurant / Indian-Buffet / Infinite-*

*We meet again on the 22.10.14 to register student pairs and assign topics.*

### Update:

Teams and topics have been selected. We are expecting teams to report back to us once per month (each team should schedule an appointment, on some Wednesday, in November and one in December) so that we can supervise progress.

Important dates:

- 14th January: Deep Learning (speakers: S. Cao and Q. Lu). slides video
- 21st January: Non-parametrics (speakers: H. Zaheri and P. Khoshdani). slides video
- 28th January: Active Learning (speakers: M. Delgado Borda and S. Anwer). slides video
- 4th February: Chinese Restaurant and Indian Buffet Process (speaker: S. Peresandra). slides video
- 4th February: Markov-Chain Monte-Carlo (speaker: S. Meusel). slides video
*11th February: test on all topics.**27th February: Report deadline.*