A0383
Title: Accelerating approximate Bayesian computation methods with Gaussian processes
Authors: Shijia Wang - Nankai University (China) [presenting]
Abstract: Approximate Bayesian computation (ABC) is a Bayesian inference algorithm class that targets problems with intractable or missing likelihood functions. It approximates the posterior distribution by utilizing simulators to draw synthetic data. However, ABC is computationally intensive for complex models in which simulating synthetic data is very expensive. An early rejection Markov chain Monte Carlo (ejMCMC) sampler is proposed with Gaussian processes to accelerate inference speed. Samples are rejected early in the first stage of the kernel using a discrepancy model, in which the discrepancy between the simulated and observed data is modelled by the Gaussian process (GP). Hence, the synthetic data are generated only if the parameter space is worth exploring. In addition, the proposed method is employed within an ABC sequential Monte Carlo (SMC) sampler. In the numerical experiments, examples of ordinary differential equations, stochastic differential equations, and delay differential equations are used to demonstrate the effectiveness of the proposed algorithm.