Simon Bartels

PostDoc @ University of Copenhagen

View My GitHub Profile

Research Interests

On the theoretical side, I am pursuing the question: how much data is necessary to obtain a decent (Gaussian process) model. Importantly, I want an answer for a dataset at hand and NOT for all datasets. Optimal stopping and probably-approximately-correct bounds are tools, I am interested to answer that question.

On the practical side, I want to improve the protein-design workflow. With Bayesian optimization, I hope to reduce the number of expensive and time consuming wetlab experiments. Using results from preceding experiments, unlabelled data from related tasks and simulations, should allow to make informed choices about which protein modifications are worthwhile to explore.

Highlighted Publications

Adaptive Cholesky Gaussian Processes (arXiv)

with Kristoffer Stensbo-Smidt, Pablo Moreno Muñoz, Wouter Boosma, Jes Frellsen and Søren Hauberg

Kernel-Matrix Determinant Estimates from stopped Cholesky Decomposition (arXiv, code)

with Wouter Boosma, Jes Frellsen and Damien Garreau

Probabilistic linear solvers: a unifying view (publication, arXiv)

with Jon Cockayne, Ilse Ipsen and Philipp Hennig