Luca Wellmeier
Nationality: German
Host: UiT The Arctic University of Norway
| Background: Luca did his bachelor’s at Bielefeld University and my master’s at the University of Padua, where he was part of the honors college Scuola Galileiana. He wrote his thesis at the Machine Learning Center Genoa (MaLGa). After that he worked as a freelance software developer and deep learning engineer. |
| Master thesis: I”High-capacity hypothesis spaces in modern statistical learning theory” The classical idea of a bias-variance trade-off of statistical learning does not seem quite valid anymore when looking at large neural networks with millions or billions of parameters: they are perfectly able to learn a whole dataset by heart while generalizing well on new examples. My thesis project summarized some of the attempts to explain that behavior (most notably the concepts of benign overfitting and double descent). I studied them via Sobolev kernels which turn out to be a great proxy for neural nets with some striking similarities but are much easier to study due to their linear nature. |
| PhD Goals: His main goal is to become a great researcher, and he thinks that the TENSORS network gives him exactly the right platform for that. He is very excited about collaborating with and learning from many different people and teams, and he wants to produce impactful contributions in the fields of real algebraic geometry and optimization. He also wants to complement his research with his experiences in statistical learning theory and deep learning wherever he can. |
| Research interests: he generally likes to tackle problems by mobilizing tools that come from multiple different fields and to make them play well with each other. Currently, these would be algebraic geometry, optimization, harmonic analysis and deep learning. |
| Hobby: He loves building things with code or with his hands, discovering new music, playing pretty much every sport that involves a ball and exploring nature, especially marine areas. |
