Gatsby Computational Neuroscience Unit
Sainsbury Wellcome Centre
25 Howland Street
London W1T 4JG UK
+44 (0)7795 291 705
I am a Reader (Associate Professor) with the Gatsby Computational Neuroscience Unit, part of the Centre for Computational Statistics and Machine Learning at UCL. A short biography.
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.
Talk slides for the NIPS 2016 workshop on generative adversarial networks (more detailed slides from the Dagstuh workshop). Adaptive MMD test paper and code; linear-time ME test paper and code.
JMLR paper on Learning Theory for Distribution Regression. Minimax rates are attainable for regression in the two-stage sampled setting (where only samples from sampled distributions are observable). Code is in the Information Theoretical Estimators toolbox.
MERLIN (Mixture Effect Recovery in Linear Networks) constructs causal variables from observed, multivariate non-causal variables as a manifold optimization problem, with application to EEG data paper and code.
NIPS oral presentation: adaptive linear-time two-sample tests, with power matching quadratic-time tests: Paper and code.
Linear time nonparametric independence tests: Paper .
A Kernel Test for Three-Variable Interactions with Random Processes, also containing much simpler proofs for the asymptotics of kernel statistics (including HSIC for basic independence testing), UAI 2016. Paper .
A kernel test of goodness-of-fit, ICML 2016. Paper , talk slides, and code.
Kernel mean shrinkage estimators, JMLR, 2016. Paper .
Filtering with State-Observation Examples via Kernel Monte Carlo Filter, Neural Computation, 2016. Paper .
A Test of Relative Similarity for Model Selection in Generative Models, in ICLR 2016 . Paper .
I co-chaired AISTATS 2016 with
Christian Robert .
This took place from 9-11 May 2016 in Cadiz, Spain.
Slides for my NIPS 2015 workshop talks are online: see the talks page.
Fast Two-Sample Testing with Analytic Representations of Probability Measures, NIPS 2015. A class of powerful nonparametric two-sample tests with a cost linear in the sample size. Code.
Gradient-free Hamiltonian Monte Carlo with Efficient Kernel Exponential Families, NIPS 2015. Adaptive Hamiltonian Monte Carlo, where the target gradient is learned from the past chain samples. Demonstrated using experimental studies on Approximate Bayesian Computation and exact-approximate MCMC. See also Heiko's blog post . Code is now online.
Slides online for the kernel course at the Machine Learning Summer School in Tuebignen. See the teaching page .
Kernel-Based Just-In-Time Learning for Passing Expectation Propagation Messages A fast, online algorithm for nonparametric learning of EP message updates (UAI 2015). Code.