Master's thesis projects

If you are enrolled in a master's program and interested in writing your thesis in collaboration with the UiT Machine Learning Group, we propose some interesting projects you can get involved with. You may also contact directly one of the group members if you wish to work on a topic that is not listed here.

Deep learning aided quantification in PET imaging

Positron emission tomography (PET) is a medical imaging technique that visualises the distribution of an injected radioactive tracer in living subjects. Medical imaging with PET plays an important role in the detection, staging, and treatment response assessment of many diseases, including cancer, neurological and cardiovascular conditions, as well as inflammation and infection.

PET is quantitative, in the sense that it allows not only visualisation but also non-invasive quantification of regional tracer uptake. Specifically, with dynamic PET imaging, it is possible to fully assess the time-dependent tracer distribution in the body. This allows quantification of biological processes, such as glucose metabolism or blood flow, by using tracer kinetic modelling. Tracer kinetic modelling, however, requires accurate determination of an arterial input-function (AIF), i.e. the tracer time-activity curve in blood.

The gold-standard AIF is obtained by measuring the time-dependent FDG radioactivity concentration in arterial blood through invasive blood sampling. This procedure is complex, time-consuming, and potentially painful with risk for complications.

The aim of this Master´s project is to develop novel methodology for arterial input function prediction, building upon state-of-the-art deep learning methods. In addition, the project will focus on developing interpretability and uncertainty methods to explain outcomes and the variability in the predictions. Both preclinical (mice) and clinical (human) PET data will be explored in the project.


  • Relevant machine learning courses, for instance FYS-2021 and FYS-3033

  • Programming skills in Python, preferably using PyTorch and/or TensorFlow

  • Experience with medical image processing is an advantage

Contact persons

Luigi Luppino (

Samuel Kuttner (

Mutual information and deep learning

One disadvantage of deep learning models is the lack of easily interpretable explanations. One way of explaining the predictions is to provide an overview of which parts of the input data are important for the prediction for example which pixels lead to the classification of ‘dog’. However, for time series data, a list of the contributing timesteps will in most cases not provide an interpretable explanation.

Mutual information can be used as a measure of the quality of internal representations in deep learning models and could provide a way to explain time series predictions.

The goal of the project is to investigate the use of mutual information as a way to identify and characterize contributing information from time series either in the time or in the frequency domain. Meanwhile, it would be interesting to extend existing information theory-based explainable AI (XAI) framework (such as information bottleneck) for time series, such that practitioners can identify the most informative frequency components or segments of time series that are most influential to the decision.

The resources, including data, that you will make available for the project(s)

We will use standard public available benchmark data for the project. The existing code base will be made available for the student (pytorch). This project is a collaboration with SINTEF Digital.

Contact person: Shujian Yu (UiT, shujian.yu@uit,no)

Graph neural networks

Graph neural networks are very powerful tools but have some limitations, like over-smoothing and over-squashing (, ). The project will explore the causes of these limitations and try to propose solutions.

Recommended prerequisites: Knowledge of machine learning and signal/image processing. Good programming skills (Python).

Contact person: Benjamin Ricaud (

Hidden Markov Model Time Series Segmentation

Hidden Markov model remains the most used model for time series segmentation. Its main advantage, discrete hidden states and observations is also its main limitation. Of course, there have been several works extending to model to more versatile distributions. Our objective will be to train a recurrent network to segment time series by enforcing sparsity of the latent representation. All that in an unsupervised manner of course. We will investigate what does the model "naturally" extract and eventually how to guide it. Possible applications include medical data, electrical usage, NLP, and many others.

Recommended prerequisites: FYS-3012, FYS-3033

Contact person: Ahcene Boubekki (

Population counting using Drone Images for Marine Surveys

Marine surveys require use of valuable resources (expert's time and boats). UiT in collaboration with Norwegian Polar Institute and University of Southern Denmark is working towards developing a solution for performing population counting based on images captured from flying a drone. The initial plan is to develop a supervised learning based methodology for detecting the number of porpoises in an image. Later on, the plan is to further develop the framework to accommodate for other similar mammal species (with fewer training samples).

Prerequisites: FYS-2021, FYS-3033

Contact person: Puneet Sharma (

Safe AI using Bayesian Deep Learning

Current decision support tools are usually designed by using expert knowledge or data driven techniques. However, these methods are mostly dependent on the high level of understanding of the subject or a dataset with unrealistic high quality to achieve optimal or desired performances. Many real-world problems are highly complex, which require new techniques that can model uncertainties and making decisions based on the availability and quality of data. Approaches toward building a personalized decision support tools include developing a prediction model of the risk and outcome, or deriving safe and effective data driven decision algorithms. With the development of artificial intelligence, deep learning has been used extensively in modelling and prediction. The combination of deep learning with Bayesian inferencing allows information and uncertainties to be accurately estimated from the training data. The AI agent needs to be designed carefully such that it can safely explore the environment and propose actions that are both risk-averse and robust. Integrating deep learning, Bayesian inferencing with reinforcement learning framework will bring great opportunities to solve the problem and contribute toward a safe AI.

Background: A background in Bayesian inference, deep learning and reinforcement learning would be ideal, but a general background in machine learning and statistical methodology will be sufficient. Good programming skills are required.

Contact persons: Fred Godtliebsen, UiT Machine Learning Group and Phuong Ngo, Norwegian Centre for E-health Research


[1] Ngo, P. and Godtliebsen, F., “Data-Driven Robust Control Using Reinforcement Learning,” 2020. [Online]. Available:

Robustness and transferability in CNNs

The issue of investigating robustness and transferability in convolutional neural networks (CNNs) has long been studied concomitantly with the success of a variety of architectures. The motivation for studying this problem lies in the fact that CNNs typically perform poorly when the testing dataset distribution significantly differs from the training set and have been shown to pick up spurious correlations in the training data (essentially getting "fooled").

This project aims to investigate mechanisms to quantify such spurious correlations and evaluate the models robustness and aims to shed more light on an extremely active area of research in deep learning.

Recommended prerequisites:

FYS-3012 Pattern Recognition, FYS-3033 Deep Learning. Good programming skills.

Contact Persons:

Michael Kampffmeyer (

Rwiddhi Chakraborty (