Currently, I’m working on the safe exploration problem in reinforcement learning. In particular, I look at ways to use verification techniques to distinguish safe exploration strategies from unsafe ones. In the long run, I’m interested in extending safe exploration to solve similar side-effects problems as you try to solve in AI alignment. I applied to the position before the CDT was announced. My supervisor suggested that I could join the CDT and I liked the idea of being part of a PhD cohort.
Experience: I was a visiting researcher at FHI in Oxford. I did my master thesis in the distributed computing lab at EPFL. Outside of research, I’m heavily involved in the effective altruism movement. Dancing lindy hop also takes up a fair deal of my time.
Visit Henrik’s public profile: http://www.imperial.ac.uk/people/c.aslund19