BSc, MSc and Pre-Thesis at MMIS Topics

We offer various interesting and challenging topics for theses in the scope of - Artificial Intelligence in general

  • Machine learning, in particular artificial neural networks
  • Human behaviour analysis
  • Assistive systems for people with cognitive and physical impairments
  • Natural language processing

    General requirements

    • You should be familiar with Latex and GIT
    • You should be familiar with working on a linux command line, also remotely via ssh
    • You need a login at the computer science department
    • You should have passed at least some of the AI-focused lectures offered by the MMIS group successfully

    How to apply?

    To apply for a thesis, please send an email to Dr. Sebastian Bader containing the following information:

    • List of MMIS AI-lectures you have attended, including the final grade achieved by you
    • Summary of your study results so far
    • Your solution to the self assessment task provided below
    • The project (one of the current research projects) you are interested to work in

    Small Self-assessment task

    Before applying for a thesis at the chair of MMIS, please solve one of the following small self assessment tasks to show your skills:
    1. Signal Analysis

    Analyse the following datafiles (SelfAssement1.csvSelfAssement5.csv). They contain a mixture of three signals each and we would like to know which signals these are and how you analysed the datasets. Please submit your solution as a python script, Jupyter notebook, or R script. Your file has to load the data file, describe your ideas while analysing the data, and finally show a description as well as a plot of the three sub-signals.

    2. Neural Networks

    Build a convolutional auto-encoder using Keras and train it using the CIFAR10 dataset. Afterwards remove the decoder part and replace it with a classification head, i.e., a fully connected sub-network which maps the latent representation generated by the autoencoder to the output classes provided as true labels from the dataset. Use only a small balanced subset of the labelled training data to train the classification head. The pre-trained encoder should help to create better classification results, compared to a classification network trained on the small labelled dataset. Explain how you created the labelled subset for training, the architectures of the two networks and the evaluation. 

      Co-Supervision for external Theses

      If you are doing your thesis in some company or research institute outside the University of Rostock, you will need a supervisor here. We are accepting co-supervision of your thesis under the following conditions:

      • You apply for a thesis as described above
      • You send us a detailed topic description as a one-page PDF which contains the general idea, a list of sub-tasks you will have to perform
      • Prior to signing the official documents we will have a joint telephone / video conference with your external supervisor and you. You will have to arrange that meeting!
      • The company has to allow a joint publication of the results