My research focuses on theoretical questions of machine learning with a particular focus on the theory of optimization and generalization of deep learning. This involves recent advances in over-parameterized networks such as the neural tangent kernel or the mean field limit, but also considers information theoretical notions such as the information bottleneck. Besides that, I am interested in all sorts of breathtaking advances in deep learning, e.g. GANs, generative language models, or beating humans in DotA.
Mark graduated from TU Dortmund with Bachelor's degrees in Mathematics (2017) and Physics (2018). After spending a year with internships at Bosch, Solon Management Consulting and Princeton University, he read Part III of the mathematical Tripos at the University of Cambridge, graduating with a Master of Advanced Studies (MASt) in 2020. Before joining us as a PhD student, Mark conducted research projects related to data analysis, machine learning and computer vision at Duke University (2017), Princeton University (2019), and the University of Cambridge (2020).
Training neural networks amounts to about 3 kg of carbon emissions per day and GPU.