Hello world! I am Søren from Denmark and welcome to my personal website. I am a researcher and engineer educated within mathematical modeling and machine learning. This website is a collection of small updates from my work life and a way to share my portfolio.
Current position
I currently work in WS Audiology as a Machine Learning Engineer in the AI Accelerator team. I work with machine learning approaches for denoising and speech enhancement.
I can’t share any more details at this point (the work is mysterious and important 😆).
Previous experience
I finished my master in the summer 2015 and shortly thereafter I started on my Ph.D. supervised by professor Morten Mørup, associate professor Mikkel N. Schmidt and associate professor Kristoffer H. Madsen. In my project, entitled Modeling Temporal Dynamics in Functional Brain Connectivity, we investigated how to quantify and qualify assumptions made when modeling temporal brain dynamics. This was achieved by using a Bayesian Hidden Markov Model framework; a probabilistic model that can track state changes in various systems. By parameterizing the dynamics in different ways, each encoding specific assumptions, and applying them to held out data, we were able to show which models had the best predictive power. I handed in and defended my thesis in fall of 2018.
After my Ph.D., I worked briefly as a postdoc at Copenhagen University, supervised by professor Kamilla W. Miskowiak. We investigated the use of machine learning for predicting treatment status based on brain imaging data in a cohort of patients with mood-disorders.
In the spring of 2019, I transitioned to the industry, working as a digital signal processing engineer in Sennheiser Communications (now called EPOS); a company that delivers audio solutions for office professionals (headsets, speakerphones etc.). I worked with product development, implementation of signal processing algorithms in embedded system and later on with more research oriented task and development of new algorithms. Our groups focus was to enhance the users own voice before transmission to the far-end using modern digital signal processing influenced by machine learning. I contributed to products like the EPOS Adapt 660, EPOS H3 Pro and EPOS Impact 1000 series.
From 2023 to 2025, I have been back at DTU working as a postdoc with Mikkel N. Schmidt. The project I was hired in was called machine learning-enabled fiber optic communication (MARBLE) funded by the Villum foundation and lead by principal investigators associate professor Mikkel N. Schmidt (DTU Compute) and professor Darko Zibar (DTU Electro). We investigated how modern signal processing techniques inspired from the machine learning literature can aid optical communication systems. We published two papers during that time — one on end-to-end learning of finite impulse response filters in communication channels and one on non-linear blind receiver side equalization using variational autoencoders.