Master Thesis Project – Exploring deep learning reconstruction

Master Thesis Project – Exploring deep learning reconstruction within multi-touch application

What we do
FlatFrog is a small company that solves big problems, such as optimizing the hybrid workplace. You know those old whiteboards that people have stopped using because half of their colleagues work remotely? Well, we have reinvented that whiteboard into a digital solution - FlatFrog Team Tablet - that can be used to collaborate both in physical meeting rooms and remotely. We are now looking for two master thesis students with strong interest for deep learning to do an exciting master thesis project with us.

We are looking for
A group of two thesis workers to perform a joint thesis.
Students pursuing a M.Sc. within Engineering Physics, Engineering Mathematics, Computer Science, Electrical Engineering or similar.
A group with strong interest for deep learning that possess the know-how to organize/implement/analyze different algorithms.


We offer
The chance to apply your academic expertise in a practical, real-world environment.
Ongoing guidance and feedback from a team of engaged engineers and academic professionals.
Compensation upon the completion and publication of your thesis.



Background of project
In recent years deep learning has been used to enhance tomographic reconstruction algorithms achieving state-of-the-art performance within the field of medical tomographic imaging. While most of these deep learning algorithms are general in the sense that they can be applied to tomographic reconstruction problems outside the domain of medicine, they are often designed in a way that makes deployment in other domains infeasible w.r.t. computational complexity and model accuracy.

About the project
In this thesis we would like to explore how the ideas/algorithms developed for medical tomographic imaging can be adapted to suit FlatFrogs non-medical application. More specifically we would like enhance our tomographic reconstruction results by leveraging deep learning as a post processing step, similar to what was first shown in [1]. This is expected to be a challenging problem since some key assumptions made in [1] (and most other research on the topic) does not hold true for our application
While we have a clear idea of what we would like to explore, we believe that there will be plenty of opportunity for the thesis workers to adapt the project to align with their academic interests e.g. by deep diving in the mathematical modelling challenges of the problem, or exploring how their deep learning models can be benchmarked/deployed practically. Moreover, FlatFrog is filled with engaged engineers who happily will discuss difficult problems that may come up during your thesis work.

In this thesis, you will
Conduct a literature study on deep learning based tomographic reconstruction.
Implement state-of-the-art deep learning-based post processing for enhancing tomographic
reconstructions.
Plan, implement and evaluate improvements.
A more detailed description of the project will be shared in due course.


Start date
Januari 2024, or earlier if possible.


Application & contact
Don’t hesitate to contact us for questions about the project!
If you find this exciting, please send your CV and short motivational letter to Amanda.wiberg@flatfrog.com. Be sure to include why you find this project interesting and how your studies match this master thesis project. We look forward to hearing from you!


[1] K. H. Jin, M. T. McCann, E. Froustey and M. Unser, "Deep Convolutional Neural Network for Inverse Problems in Imaging," in IEEE Transactions on Image Processing, vol. 26, no. 9, pp. 4509-4522, Sept. 2017, doi: 10.1109/TIP.2017.2713099.