Approximate Bayesian computation: a simulation based approach to inference

There is a large class of stochastic models for which we can simulate observations from the model, but for which the likelihood function is unknown. Without knowledge of the likelihood function standard inference techniques such as Markov Chain Monte Carlo are impossible, as the unnormalized likelihood function is explicitly required for the calculation of an acceptance rate. In this talk I shall introduce a group of Monte Carlo methods that can be used to perform inference for stochastic models from which we can cheaply simulate observations.