About Me

Frederick Altrock

PhD researcher at University of Münster specializing in probability theory and applying it to neural network training dynamics and Initialization Schemes.

Who am I?

As a researcher, it’s extremely important to me that my work is not just a game of thought. I don’t spend time on problems that (in my view) have no practical relevance. My work focuses on training dynamics of ultra wide neural networks NTK and $\mu P$. The practical side of this is commonly referred to as rich regime, lazy regime and Grokking which is the transition between the two during training. The latter are often times considered from mostly a practical standpoint with lightweight mathematical theory. To me they are all closely related.

Outside of research, I enjoy what most people would call “hacking”: reverse engineering, taking systems apart, figuring out how they really work. I actually considered studying IT security, but in the end math won, and “hacking” became my hobby.

I also spend an unreasonable amount of time tuning my 3D printer and building random things. 3D printing is fun, but more importantly it makes it possible to prototype ideas fast. I wouldn’t want to give that up. Thus you will find a few posts on this too.