Machine Learning for Structured Models

Dmitrij Schlesinger, Sommer semester 2017

A common opinion is that the main task of Machine Learning is to establish a connection between raw data and their semantically meaningful interpretations. A bit more poetic: "to teach computers to think" (or at least to understand the data). Machine Learning approaches have numerous applications in many different subject areas, especially in those where data uncertainty plays a crucial role, like e.g. Natural Language Processing, Computer Vision etc. This lecture focuses on structured models, where the estimated quantity has a complex structure, i.e. it is not just a number or a class but e.g. a graph or a sequence or a grid, whatever.

Lectures: Friday, 3DS, 11:10-12:40, INF E023, Start: 07.04.2017
Exercises: Monday, 4DS, 13:00-14:30, INF E007 and Thursday, 2DS, 9:20-10:50, INF E007, Start: 24.04 and 27.04 respectively
Prerequisites: Solid mathematical background. This lecture is especially suitable for people, who attended Machine Learning I and like to deepen their knowlege. Of course, the lecture is open to everyone, but if you did not attend "Machine Learning I", be ready for self-study.
Extent: 2/2/0, Exam: Oral exam after the semester, Enrollment: jExam, if does not work – email, Maximum attendees: 60
Note: The lectures will be held in English.

News:
28.06.2017, Exam info: Im principle, I am available 24.07-30.09 and I am quite flexible in that time. Hence, if you plan an exam with another examinator (e.g. ML1+CV1, whatever), ask the other examinator first (with Cc: to me for any case). For exams with me only (i.e. ML1 only, or MLfStruct only, or ML1+MLfStruct) my preferred exam week is 7.08-11.08 due to some organisational reasons. Just write me mail, state explicitly, whether you can or you can not in 7.08-11.08. Please care by yourself about all necessary forms (protokols, etc.) and just bring them to the exam.
08.06.2017: Room changes: The exercise on 15.06 will be in APB2026 (our lab on the second floor of the informatics building). The lecture on 23.06 will be in SCH/A216 (Georg-Schumann-Bau, close to the tram stop "M√ľnchener Platz").
29.03.2017: Welcome! On this page, all required information is provided during the semester – lecture scripts, exercise assignments, some actual informations, whatever. The scripts will appear during the semester, step by step. See you at the first lecture on Friday 07.04.

Scripts:
Lectures:
07.04: Introduction, Markov Chains, (annotated)
28.04: Hidden Markov Models
05.05: Inference in HMMs, (annotated)
12.05: Maximum Likelihood for Markov Chains, (annotated)
19.05: Markov Trees
26.05: Labeling Problems, (annotated, typos corrected)
02.06: Submodular MinSum Problems
16.06: Submodular MinSum Problems (cont.), (annotated)
23.06: Search Techniques, (annotated)
30.06: LP-relaxation
07.07: MRF learning (annotated)
14.07: Summary
Exercise assignments:
24.04, 27.04: Markov Chains
04.05, 08.05: Inference in Markov Chains
11.05, 15.05: Inference in Markov Chains (cont.) + rests
18.05, 22.05: Maximum Likelihood for Markov Chains + rests
29.05, 01.06: Labeling Problems
12.06, 15.06: Binary MinSum Problems
19.06, 22.06: Submodular MinSum Problems
26.06, 29.06: Search Techniques + rests
03.07, 06.07: LP-relaxation + rests (as usual :-) )
10.07, 13.07: MRF Learning
Last touch: 13.07.2017