Master thesis projects

Classification of brain signals (EEG) with deep learning

 

Description

Brain-computer interfaces might be a major upcoming paradigm in human-computer interactions. This project aims at using deep learning and signal processing to extract features from raw EEG signals in order to predict and capture the intention, behavior, or perception of the subject. You will have to prepare and perform controlled experiments (with human subjects), collect data, and build processing and modeling pipelines, and test hypotheses.

ML techniques

  • Deep learning for signal processing and multivariate time series encoding, decoding, and classification.

References

[1] https://iopscience.iop.org/article/10.1088/1741-2552/ab260c

Please include the following in your application:

  • 3-4 sentences (email body) about what could be potential practical challenges with this project and how you’d solve them.
  • Github link
  • Grades
  • CV

Apply

Machine learning for generative models in medical applications

 

Description

Data in health care and medical applications has two main problems: 1) it is difficult and expensive to collect data and 2) there are privacy concerns in data handling and sharing. One way to solve both these issues is to generate privacy preserving synthetic data using techniques as described in [1, 2]. We are interested in applying such techniques to data we are exposed to, in order to see if they create value in our context.

ML techniques

  • Synthetic data generation (not necessarily GANs)
  • Diagnosis prediction methods

References

[1] https://towardsdatascience.com/reducing-ai-bias-with-synthetic-data-7bddc39f290d
[2] https://openreview.net/pdf?id=S1zk9iRqF7

Please include the following in your application:

  • 3-4 sentences (email body) about what could be potential practical challenges with this project and how you’d solve them.
  • Github link
  • Grades
  • CV

Apply

Machine learning for future sentiment prediction

 

Description

Public companies need to disclose vital information that affects company (stock) performance through press releases. Using NLP methods, it is possible to determine the sentiment of individual texts. We are interested in taking that to the next level: to predict the sentiment of future press releases using available written information, also considering context, e.g. what the competitors are doing.

The project focus can be on either NLP techniques to extract sentiments and other important information that effects general prediction performance, or on prediction methods.

ML techniques

  • NLP for sentiment analysis (zero shot learning BART for example)
  • Prediction methods (Bayesian? to get sentiment uncertainty)

Please include the following in your application:

  • 3-4 sentences (email body) about what could be potential practical challenges with this project and how you’d solve them.
  • Github link
  • Grades
  • CV

Apply

Machine learning for explainability in medical time series

 

Description

Explainability techniques is very much an active research area and there are some very important research questions such as what methods should be used when and for what applications. We want to investigate these open questions in the domain of medical time series classification models in order to understand which techniques are relevant for diagnostics. In particular, it is interesting to investigate how different model architectures affect the final interpretation of the prediction and the corresponding explanation, in particular, if it makes sense to be true to the model or true to the data when looking at the medical time series (see [1]).

ML techniques

  • Integrated/expected gradient methods
  • Neural time series classification models

References

[1] https://arxiv.org/abs/2006.16234

Please include the following in your application:

  • 3-4 sentences (email body) about what could be potential practical challenges with this project and how you’d solve them.
  • Github link
  • Grades
  • CV

Apply

Machine learning for company risk in ownership graphs

 

Description

Insolvency/bankruptcy risk prediction for smaller companies very related to the individual risks of the owners: their respective financial strength, their experience in running companies and the history of previous endeavors. Given a dataset with data on personal connections to different companies over time, and how risk in one company affects the risk in related companies depending on the owners, we can analyse the corresponding company/private person graph, either by using graph methods or standard ML methods, to get a better idea of the risk in each individual company.

ML techniques

* Many different alternatives are possible: graph neural networks, classification methods, state space models etc.

Please include the following in your application:

  • 3-4 sentences (email body) about what could be potential practical challenges with this project and how you’d solve them.
  • Github link
  • Grades
  • CV

Apply

Denna sida använder cookies. För mer information kan du läsa om cookies här.