Output details
11 - Computer Science and Informatics
University College London
Kernel Bayes’ Rule
<13>Much Bayesian inference requires approximating the posterior for intractable models: including variational methods and sampling approaches, hundreds of papers and multiple books have been written, often requiring significant work on problem-specific implementations. We achieve provably consistent Bayesian inference on models learned entirely from data, without problem-specific engineering, which is especially useful when no good models are known (we recently applied to branching processes for population genetics). A version is under review with JMLR. A number of invited talks and tutorials covered this work, including at APRM2012 and the ML summer school in Kyoto(2012), CMU, LSE Statistics, Oxford Statistics, and INRIA.