This talk will present an overview of my research. I would like to use this talk and visit to identify problems of mutual interest to MSR UK researchers and my new team at Oxford CS. The talk will cover the following five topics: (i) Bayesian optimization and bandits for automatic machine learning, analytics, personalization, control and algorithm configuration. (ii) Deterministic Gibbs samplers and de-randomization. (iii) From exponential to linear complexity when learning some practical MRFs. (iv) Bayesian deep learning. (v) Theoretical analysis of batch and online random forests, and the development of new ensembles.