links
Gatsby Computational Neuroscience Unit CSML University College London
Collaborations
• I am afflilated with the MPI for Biological Cybernetics as a research scientist.
• I am in collaboration with the Select Lab at CMU.
Contact
arthur.gretton@gmail.com
Gatsby Computational Neuroscience Unit
Alexandra House
17 Queen Square
London - WC1N 3AR
Phone
207-679 1186
info
I am a lecturer with the Gatsby Computational Neuroscience Unit, part of the
Centre for Computational Statistics and Machine Learning
at UCL.
My current research focus is on using kernel methods to reveal properties and relations in data. A first application is in measuring distances between probability distributions. These distances can be used to determine strength of dependence, for example in measuring how strongly two bodies of text in different languages are related; testing for similarities in two datasets, which can be used in attribute matching for databases (that is, automatically finding which fields of two databases correspond); and testing for conditional dependence, which is useful in detecting redundant variables that carry no additional predictive information, given the variables already observed. I am also working on applications of kernel methods to inference in graphical models, where the relations between variables are learned directly from training data: applications include cross-language document retrieval, depth prediction from still images, and protein configuration prediction.
Recent news
• ICML 2012 workshop on RKHS and kernel-based methods: theoretical topics and recent advances (with Zaid Harchaoui and Bharath Sriperumbudur)
Older news
• A Kernel Two-Sample Test , JMLR, March 2012. The
Matlab software is also updated.
•
Lecture slides
for Advanced Topics in Machine Learning (COMPGI13), and for a short course on kernels
•
Talk slides
on kernel distribution embeddings, including applications in hypothesis testing and inference (at Oxford statistics department)
•
Kernel Bayes' Rule
NIPS 2011 (see also the
technical report with proofs)
•
Modeling transition dynamics in MDPs with RKHS embeddings of conditional distributions, on using kernel conditional mean embeddings for learning the value function and optimal policy in MDPs
