Accelerating ABC using emulation and history-matching Seminar

Approximate Bayesian computation (ABC) algorithms are Monte Carlo algorithms that can be used to do Bayesian inference for stochastic models without explicit knowledge of the likelihood function, and in the past decade they have become very popular, particularly in the biological sciences. In this talk I\'ll describe the basic ABC approach, explaining how I believe we should view ABC algorithms, and draw links between ABC and history-matching. Finally, I\'ll describe a new method for using Gaussian process emulators to speed up ABC algorithms by approximating the likelihood function, based on the synthetic likelihood approach proposed by Wood (2010).