course: Master project Pattern recognition

teaching methods:
responsible person:
Prof. Dr.-Ing. Do­ro­thea Kolossa
Prof. Dr.-Ing. Do­ro­thea Kolossa (ETIT)
offered in:
winter term

dates in winter term

  • kick-off meeting: Friday the 30.10.2020 from 10:00 in Online


Die Angaben zu den Prüfungsmodalitäten (im WiSe 2020/2021 | SoSe 2021) erfolgen vorbehaltlich der aktuellen Situation. Notwendige Änderungen aufgrund universitärer Vorgaben werden zeitnah bekanntgegeben.
Form of exam:project
Registration for exam:Directly with the lecturer
continual assessment


Also this semester, modern methods for time series analysis, mainly structured hybrid models ([1], [2]), will be the subject of the practical course. The practical part of the course includes the creation of training data and the implementation of the model by the project participants themselves. Details on the generation of the training data will be announced at the first appointment (Formula One).

The following literature list is recommended for preparation:

[1] Michael C. Burkhart, David M. Brandman, Carlos E. Vargas-Irwin, Matthew T. Harrison, "The discriminative Kalman filter for nonlinear and non-Gaussian sequential Bayesian filtering"

[2] Matthew J. Johnson, David K. Duvenaud, Alex Wiltschko, Ryan P. Adams, Sandeep R. Datta "Composing graphical models with neural networks for structured representations and fast inference", NIPS 2016

or as a video

During the semester, there will be a weekly online meeting (with compulsory attendance) to discuss the progress of each week and plan the next steps. The lab will be concluded by a written report documenting the submitted code and results, and an online final presentation.


The focus of this course in winter semester 2020/21 is set on practical solutions to machine learning problems. Students implement a joint project in teams of 2 to 3 members over the course of a semester.

In the winter semester 20/21 attendance is not possible. However, we would like to enable students to deal with current research topics both theoretically and practically within the framework of this master's project.



recommended knowledge

  • Basic knowledge of digital signal processing and machine learning
  • confident command of at least one programming language
  • State space models (Kalman filter, Hidden Markov models)
  • artificial neural networks and bayesian methods (Variational Autoencoder)



Interested students must register for the lab by sending an email (to The deadline for registration is October 29.