Papers
5
Total Citations
134
H-Index
4
About
Arthur Gretton is a leading figure in machine learning, renowned for pioneering work in kernel methods, statistical inference, and nonparametric hypothesis testing. His research fundamentally reshaped how we model complex, high-dimensional data, moving beyond traditional parametric assumptions. Gretton’s major contributions include developing the theory of kernel embeddings of distributions, a powerful framework that represents probability distributions as elements in a reproducing kernel Hilbert space. This work enabled a suite of novel algorithms for two-sample testing (e.g., the Maximum Mean Discrepancy) and independence testing (e.g., the Hilbert-Schmidt Independence Criterion), each garnering thousands of citations and becoming standard tools in the field. His highly cited early work on support vector regression for system identification (over 90 citations) laid the groundwork for later breakthroughs in kernel Bayesian inference, including the development of the Kernel Monte Carlo filter for state-space models. Gretton’s influence extends to robotics and human action recognition, where his characteristic kernels on structured domains have proven exceptionally effective. He is also a co-author of the foundational "Model-based kernel sum rule," advancing principled nonparametric inference in probabilistic graphical models.
Research Focus
Key Achievements
Top Papers
- 1Support vector regression for black-box system identification91 citations · 2002
- 2Filtering with State-Observation Examples via Kernel Monte Carlo Filter16 citations · 2015
- 3
- 4Monte Carlo Filtering Using Kernel Embedding of Distributions12 citations · 2014
- 5