Giant leap in ABC inference scalability

HIIT scientists Michael Gutmann and Corander published a machine learning based ABC inference approach in the Journal of Machine Learning Research. Their method (BOLFI) is based on Bayesian optimization with Gaussian processes and is generally applicable to simulator models with intractable likelihoods. Without sacrificing accuracy, BOLFI speeds up posterior computation by 3-4 orders of magnitude compared with the state-of-the-art sequential Monte Carlo algorithms. It is expected to become a new standard in ABC inference, paving way for a multitude of new applications where the earlier methods have been too expensive computationally. Details of the method can be found in the article: Bayesian Optimization for Likelihood-Free Inference of Simulator-Based Statistical Models. Michael U. Gutmann, Jukka Corander; 17(125):1−47, 2016. http://jmlr.csail.mit.edu/papers/v17/15-017.html