My name is T. Konstantin Rusch and I’m a PhD student in applied mathematics at ETH Zurich,
supervised by Siddhartha Mishra .
My research focuses on physics-based machine learning, where I’m interested in both, using physical priors to construct new methods which can be applied in classical machine learning domains (e.g. time-series prediction, speech recognition or NLP), as well as leveraging machine learning methods to solve scientific problems (e.g. solving physical systems modeled by PDEs). The goal of my research is to combine both directions by constructing new state-of-the-art machine learning methods using physical priors to solve scientific problems.
Before I started my PhD at ETH Zurich, I obtained my Master’s degree in computational applied mathematics at The University of Edinburgh, and before that I finished my Bachelor’s degree in mathematics at the University of Bonn.
Contact: konstantin.rusch [at] sam.math.ethz.ch
|Oct, 2021||New preprint (joint work with UC Berkeley) Long Expressive Memory for Sequence Modeling : A novel method for sequence modeling based on multiscale ODEs that is provably able to learn very long-term dependencies while being sufficiently expressive to outperform state-of-the-art recurrent sequence models.|
|Aug, 2021||The recording of my spotlight talk at ICML 2021 about our UnICORNN paper is now publicly available. If you want, you can watch it here .|
|May, 2021||I recently got interviewed by the famous machine learning/AI podcast TWIML AI. I spoke about our research on using physics (specifically Hamiltonian systems and systems of coupled oscillators) to construct novel state-of-the-art RNNs. If you want, you can check it out here .|