On this page I show some of the cool projects I made over the years.
In my master's thesis, I demonstrate how Concept Bottleneck models can produce deceptive explainability, meaning that they first make a prediction and then come up with an explanation to justify the prediction, without the explanation telling anything about which concepts are present.
In my bachelor's thesis, I show that Archetypal Analysis can be used on data from the European Social Survey to recreate Schwartz's theory of basic human values as archetypes. The thesis also describes a method to perform Archetypal Analysis on ordinal data. While my method failed to scale properly to a new scale, others were able to build on my work and publish it.
The first project I made in my bachelor's was to use transfer learning to predict what emotions an image would evoke in humans. Note: PDF in Danish
Does a robot expressing emotions change its ability to influence humans in a gaze cueing task? We found no significant effects. But I learned to perform Cognitive science experiments.
Like everyone else, I proved that I know Reinforcement Learning by implementing a classical project from Sutton and Barto.
Can you be an ethical investor and still make money? In this portfolio optimisation project I include ESG scores in a Markowitz portfolio optimisation. The most interesting part was to see an algorithm perform utilitarian calculations.
This project tries to use pre-trained CNNs to map outlier EEG signals for the company BrainCapture.
A small social graph project where we explore which MBTI personality types are on Reddit, which types talk together, and the sentiment of the conversations.