About – Kai Ueltzhöffer

Kai_Klein
Kai Ueltzhöffer

Right now I’m a full-time medical student at the University of Heidelberg, Germany. I also studied Physics there, ages ago, which I finished with a diploma thesis (indeed, I finished before the bachelor/master system was widely adopted in Germany) on image processing for studying the microstructure of polar ice cores (the only climate archive from which you can get actual samples of the past atmosphere).

After that I took a short swing into Astronomy, searching for extrasolar planets with small robotic telescopes using consumer hardware. But soon I got hooked by the human brain and started a PhD in Physics (but rather Cognitive and Computational Neuroscience) at the University of Frankfurt am Main. Here I implemented a very simple model of working-memory (sort of the RAM of our brain, only that it also can do some computation by itself) and decision-making using recurrent neural networks, which could be fitted to behavioural data of individual subjects doing a computer experiment in which they had to switch between two simple tasks. Somehow this model was even physiological enough to predict the functional MRI response of the individual subjects in brain areas well known to be relevant for task-switching. During the second year of the PhD, however, I had the feeling that studying the brain without a good overview of the basic anatomy, biochemistry and physiology of the human body would not get me too far (c.f. the idea of embodiment), so I started studying medicine. Last year (October 2016) I finally defended my Physics PhD and now I’m mostly focused on studying medicine and doing my MD thesis on reactive aggression in borderline personality disorder at the Psychiatry Department here at Heidelberg.

In my spare time I like to play at the interface of machine learning, mainly variational inference and sampling methods in combination with deep (recurrent) neural networks (such as the great work by Kingma & Welling, Rezende, Mohamed & Wierstra, Chung & al.), and current theories in computational neuroscience (mainly the active inference framework). And this is what this blog is mainly supposed to be about. But we’ll see…