Advanced Seminar Music Matching & Scalable Machine Learning (IN2107, IN4427)

Lecturer (assistant)
Duration2 SWS
TermSommersemester 2021
Language of instructionEnglish
Position within curriculaSee TUMonline
DatesSee TUMonline


  • 12.04.2021 14:00-16:00 Online: Videokonferenz / Zoom etc.
  • 19.04.2021 14:00-16:00 Online: Videokonferenz / Zoom etc.
  • 26.04.2021 14:00-16:00 Online: Videokonferenz / Zoom etc.
  • 03.05.2021 14:00-16:00 Online: Videokonferenz / Zoom etc.
  • 10.05.2021 14:00-16:00 Online: Videokonferenz / Zoom etc.
  • 17.05.2021 14:00-16:00 Online: Videokonferenz / Zoom etc.
  • 31.05.2021 14:00-16:00 Online: Videokonferenz / Zoom etc.
  • 07.06.2021 14:00-16:00 Online: Videokonferenz / Zoom etc.
  • 14.06.2021 14:00-16:00 Online: Videokonferenz / Zoom etc.
  • 21.06.2021 14:00-16:00 Online: Videokonferenz / Zoom etc.
  • 28.06.2021 14:00-16:00 Online: Videokonferenz / Zoom etc.
  • 05.07.2021 14:00-16:00 Online: Videokonferenz / Zoom etc.
  • 12.07.2021 14:00-16:00 Online: Videokonferenz / Zoom etc.

Admission information

See TUMonline
Note: - Presentations and discussions - Written technical report (5-8 pages ACM proceedings style), to be submitted within 2 weeks after the presentation


Modulkatalog: IN2107


Two pieces of music can be similar to each other partially or fully. A varied range of audio features is used for music matching and the techniques utilised for it are ever evolving. It has varied applications like audio classification, audio retrieval, audio segmentation, speech recognition, singer identification, and music recommendation. Machine learning proves helpful in solving many such problems. The general improvement of these machine learning models is not only based on more advanced architectures or algorithms, but also on their increased quality and quantity of training data. A thorough analysis of the impact datasets, as well as their preprocessing techniques have, is desperately needed to provide a generally applicable knowledge base. Reproducibility and general robustness is an oftentimes overlooked and hidden attribute of the model training process. A formalization of the current state-of-the-art would prove beneficial for every machine learning practitioner. This seminar includes various topics around audio processing for music matching as well as the importance of data involving machine learning. Students will be assigned one topic each based on their interests. Starting from the provided literature, they would have to explore more about their assigned topics. Students will be able to make their own contributions in it. We plan for two individual meetings, one after the initial literature research, as well as one to further deepen the focus of the research. You can optionally send a short motivation statement why you want to participate in this seminar via Email to rinita.roy[AT], alex.isenko[AT] (max. 300 words) (sending a motivation statement is optional, but may increase your chances of getting a place). Please pick 3 topics which interest you from the following list and send them via Email to rinita.roy[AT] and alex.isenko[AT] We do not guarantee that you will get your proposed topic, but we’re open for suggestions. Possible topics are described here: usp=sharing Grades are going to be defined based on your research and your presentation. The focus of this seminar is to make your research presentable, as this is a key skill for each graduate student. A well done research with a bad presentation is useless and vice-versa. You may be able to get a bonus to your grade (0.3/0.4) as long as you show up for at least 3 presentation days and ask at least one question which borders on your topic and the one presenting. Preliminary meeting is on the 04.02.2021 at 11:00 at the following link: Code: 796693 Presentation link: Presentation download is under "Course Documents" in TUMOnline.


Intermediate knowledge and interest in machine learning topics and/or music are desirable. This is a Master student seminar.

Teaching and learning methods

- Presentations and discussions - Written technical report (5-8 pages ACM proceedings style), to be submitted within 2 weeks after the presentation


Grade is based on the report and the presentation

Recommended literature

T.b.d. Based on the topics