I am performing research in machine learning, and particularly by means of deep learning, for computer vision and image processing. The research covers both applied problems within computer graphics and medical diagnosis, as well as more fundamental problems in visualization and understanding of deep learning.
My background is within computer graphics and image processing. I finished my PhD studies in 2018, presenting a thesis that treats different aspects of high dynamic range (HDR) imaging. I have done research in both display and distribution of HDR video, as well as machine learning for reconstructing HDR images from conventionally photographed images.
Visualization and understanding of neural networks
Neural networks are represented by a large quantity of optimized weights. To understand and visualize these, we explore and compare a large number of trained networks. We investigate how different techniques for optimization impact the weights, and how the weights can be compared in order to understand differences between trained networks. The goal is both to gain a better understanding for how neural networks operate, but also to visualize and compare differences between different models.
Generative AI in medicine
In medical AI, one of the main limiting factors is the availability of data. It is both expensive and prohibitively time consuming to gather and annotate data, and the process relies on busy experts. Moreover, it is problematic to make data available, due to its private and sensitive nature. We investigate techniques for using generative machine learning to create large quantities of synthetic data that reflects real data, but which do not infringe on its anonymity.