About

Hi and welcome to my (Simon Schrodi’s) personal website!

About me

I am a 5th year PhD student at the University of Freiburg, working in the area of explainable machine learning advised by Prof. Thomas Brox. My research lies at the intersection of the science of deep learning, interpretability, and AI safety. In particular, I investigate how AI models learn and generalize, and how this shapes their behavior using controlled experiments and mechanistic analysis. My broader interests also include machine learning for climate science and automated machine learning. You can find some of my works under the Publications page or on my Google Scholar profile.

I am also a research scholar at MATS 9, mentored by Alex Cloud, Cem Anil, and Arthur Conmy. My work there examines the limitations of pretraining safety methods and how pretraining shapes models’ propensity for agentic misalignment.

My academic background

I started my academic journey with a BEng in Computer Engineering (1.2, equal to A) at DHBW Karlsruhe as part of a dual study program at SICK AG. At SICK AG, I worked on various deep learning solutions for their sensor systems used in the logistics industry. Afterwards, I did my MSc (summa cum laude, 1.0, equal to A) in Computer Science at the University of Freiburg. During my master’s studies, I did research on the robustness of optical flow networks (see here) and automated machine learning (see here and here). I also participated in the CVPR 2021 - NAS Competition Track 3, where we won 2nd place (see here).

Me, personally

When not coding, I like to make music and stay active. I play accordion in a local orchestra and occasionally in project orchestras. I also like playing solo, but more as a hobby. Besides that I like to go biking and hiking.