Arthur Gretton
As of October 2026, Arthur Gretton is a professor at the Gatsby Computational Neuroscience Unit at University College London, where his work on kernel methods focuses on statistical tests that check whether two samples come from different distributions. He studied physics and systems engineering at the Australian National University before completing a PhD at Cambridge. From September 2002 he worked as a research scientist at the Max Planck Institute for Biological Cybernetics, and in February 2009 he became a project scientist at Carnegie Mellon University. His 2012 paper, 'A Kernel Two-Sample Test', was written with four co-authors, and in 2017 NeurIPS gave a best paper award to 'A Linear-Time Kernel Goodness-of-Fit Test', which he co-wrote. He is also a research scientist at Google DeepMind and unit director of ELLIS Unit London.
Timeline
-
1996–2003
Degrees and a Cambridge PhD
He receives degrees in physics and systems engineering from the Australian National University in 1996 and 1998. From 1999 to 2003 he studies machine learning at Cambridge with the Signal Processing and Communications Laboratory and Microsoft Research, and he completes his PhD there.
-
Since 2002
Research scientist at the Max Planck Institute
From September 2002 he is affiliated as a research scientist with the Max Planck Institute for Biological Cybernetics.
-
February 2009
Project scientist at Carnegie Mellon
In February 2009 he becomes a project scientist in the Machine Learning Department at Carnegie Mellon University.
-
2012
A kernel two-sample test
He and four co-authors publish 'A Kernel Two-Sample Test' in 2012. The paper constructs statistical tests to determine whether two samples are drawn from different distributions.
-
2017
Best paper at NeurIPS 2017
NeurIPS gives its 2017 best paper award to 'A Linear-Time Kernel Goodness-of-Fit Test', which he co-writes with Wittawat Jitkrittum, Wenkai Xu, Zoltan Szabo and Kenji Fukumizu.
-
2020
Deep kernels for two-sample tests
In 2020 he co-writes 'Learning Deep Kernels for Non-Parametric Two-Sample Tests', a paper in the proceedings of the 37th International Conference on Machine Learning. Its tests are built from kernels parameterized by deep neural nets, trained to maximize test power.
-
December 2024
A lecture on deep proxy learning
He gives a distinguished lecture at the University of Stuttgart in December 2024, titled 'Learning to Act in Noisy Contexts using Deep Proxy Learning'.
-
As of October 2026
Professor at UCL's Gatsby unit
He is a professor at the Gatsby Computational Neuroscience Unit and director of the Centre for Computational Statistics and Machine Learning at UCL. He is also a research scientist at Google DeepMind and unit director of ELLIS Unit London.
Sources and credits
Photo: gatsby.ucl.ac.uk; Arthur Gretton; creator not stated. License: Not stated by source. Source: gatsby.ucl.ac.uk. Cropped to a square and resized. Ask us to remove or replace this photo.