Type: Master Thesis (MSc)

 

Objective  / Ziel 

This Thesis should support ongoing research at the Ubiquitous Computing Laboratory (UC-Lab) in the field of biomedical signal processing and pattern recognition using deep neural networks. The overall goal is an overview and comparison of different methods applying deep neural networks for (medical) times series classification. The networks will be used in the scope of sleep research. The result will be a proper comparison scheme for different network structures. Comparison metrics could be the capacity (size) and performance of a given structure for a given task. The comparison should pay particular attention to the special requirements in a medical signal processing context.  

 

Description / Beschreibung

In recent years, the use of neural networks has (NN) increased enormously in the areas of biomedical pattern recognition. They have shown to be effective in numerous applications, like segmentation of MRI pictures and detection sleep phases. Convolutional neural networks in particular (CNN) helped to improve pattern recognition by learning features of pattern by themselves. Due to the enormous increase in applications of neural networks, the applied networks are extremely divergent. To conduct research in this growing field, it is extremely important to gain a good overview of the methods used. Your research will play an important role in kickstarting the application of neural networks within the UC-Lab. Normal comparison of NN in solely methodical publications is mostly based on performance in terms of accuracy or speed. For our comparison it is important to take more features of these networks into account. Our requirements are based on the medical application of these networks. Examples for these features, on top of the ones already mentioned, are explainability, required dataset size for training and transferability from one problem to another. Further features and a ranking in terms of these are open for discussion and part of the thesis.

 

Task / Aufgabe

  • A systematic literature search for neural networks in the scope of (medical) signal processing
  • The development of comparison metric for neural networks 
  • Application of selected networks on a given task (sleep stage scoring)
  • Comparison of the networks due to the selected features

 

Requirements / Voraussetzungen:

Keep in mind that the following is a list of the ideal skills, the most important requirement is the motivation to work in the described research field. We will adapt the tasks considering your skillset as far as possible. Some listed skills can and shall be acquired during the thesis. To excel in your research on the given task the following skill would be great:

  • Experience in the field of signal processing
  • Experience with Python (especially with frameworks like Numpy, PyTorch or Tensorflow)
  • Experience in the field of deep learning/machine learning

The work plan can be adapted to the scientific degree's level. Appropriate hardware for the application of neural networks is available. Please feel free to contact me if you have any questions.

 

Contact

Lucas Weber, Email: This email address is being protected from spambots. You need JavaScript enabled to view it., Phone: +49 7531 206 227; Office: F-126

 

Link

https://uc-lab.in.htwg-konstanz.de/thesisteamproject/517-dnn.html

 

 

 

Events

No upcoming event!