Mark van der Wilk

Mark van der Wilk is a machine learning researcher who works on Gaussian processes and on methods that let neural networks learn their own symmetries and architecture. After a Cambridge PhD with Carl Rasmussen in 2017, he coauthored the GPflow library and the Convolutional Gaussian Processes paper, then worked in industry on data-driven decision-making. In 2019, a paper he wrote with David Burt and Rasmussen on sparse variational Gaussian process regression received an Outstanding Paper award at ICML, and in 2020 he started a research group at Imperial College London. He moved to Oxford in 2023. He is an Associate Professor in the Department of Computer Science there and a Tutorial Fellow at Hertford College, where his group works on generalisation beyond the training set and on neurons that assemble their own architecture.

Now at: Hertford College

Timeline

By 2017

A Cambridge PhD with Rasmussen

He moves from the Netherlands to the UK for an undergraduate degree in Engineering. He then does his PhD in the Machine Learning Group at the University of Cambridge with Carl Rasmussen, completing his thesis in 2017.
2017

GPflow and convolutional Gaussian processes

He is one of eight authors of GPflow, a Gaussian process library that uses TensorFlow for its core computations, described in a 2017 paper. In September 2017, he, Carl Edward Rasmussen and James Hensman post Convolutional Gaussian Processes, which brings convolutional structure into Gaussian processes for high-dimensional inputs like images.
2017–2020

Decision-making work in industry

From 2017 to 2020, he works in industry on automated data-driven decision-making problems, collaborating with Dr James Hensman.
2018

Learning invariances by marginal likelihood

In August 2018, he, Matthias Bauer, ST John and James Hensman post Learning Invariances using the Marginal Likelihood. The paper builds invariances into the model structure and learns them by maximising the marginal likelihood, with a variational inference scheme for Gaussian processes.
2019

An ICML outstanding paper

Rates of Convergence for Sparse Variational Gaussian Process Regression, written with David Burt and Carl Edward Rasmussen, is named an Outstanding Paper at ICML 2019. It shows that, with high probability, the KL divergence of the approximation can be made arbitrarily small by growing the number of inducing variables more slowly than the dataset size.
2020

A research group at Imperial

In 2020, he returns to academia to start a research group at Imperial College London, focusing on training methods for deep neural networks that make predictions more robust, adaptable, and automatic. He serves there as an assistant professor and as an associate professor.
2023

Associate Professor at Oxford

In 2023, he is appointed Associate Professor in the Department of Computer Science at the University of Oxford and Tutorial Fellow at Hertford College. In October 2023, he coauthors Learning Layer-wise Equivariances Automatically using Gradients with Tycho van der Ouderaa and Alexander Immer, which learns the amount of equivariance in layers by optimising the marginal likelihood.
As of October 2026

Oxford group on generalisation

He is an Associate Professor in the Department of Computer Science at the University of Oxford, a Tutorial Fellow at Hertford College, and runs a research group on machine learning. The group works on finding general patterns that allow generalisation beyond the training set without humans manually encoding them, and on neurons that assemble their own architecture while keeping computational costs low.