Dr. Simon Weinzierl
Postdoc
Department of Mathematics
Office address:
Theresienstr. 39
Room B445a
80333 Munich
About me
LMU Munich has been my home since 2011. The subjects I have been able to study here range from Theoretical Physics to Computer Science to pure Mathematics. After a short stint outside academia with Menlo Systems, LMU and I rekindled our relationship. Now I am working in the group of Prof. Rauhut on Linear Neural Networks and the Implicit Bias of Gradient Descent. In my free time, I like to go cycling, hiking, bouldering, and baking cakes.
Keywords
Linear Neural Networks | Implicit Bias of Gradient Descent | Algebraic Geometry
Short Biography
- Software Engineer, Menlo Systems (2023-2025)
- PhD in Mathematics, LMU Munich under Prof. Fabien Morel (2018-2024) — [dissertation link]
- Master's Degree, TMP Programme, LMU Munich (2015-2018)
- Bachelor's Degree in Computer Science, LMU Munich (2012-2015)
- Bachelor's Degree in Physics, LMU Munich (2011-2014)
Research Topics
Linear Neural Networks are neural networks with zero bias and the identity function as the activation function. Essentially, they factor the linear objective function. I am interested in applying Linear Neural Networks to matrix recovery problems such as Quantum Tomography.
It has been observed empirically that gradient descent methods used in matrix recovery problems approximated by Linear Neural Networks converge to the original signal with fewer measurements when the network has more layers. I am working on explaining this phenomenon theoretically.