Approximate Bayesian Computation based on Progressive Correction of Gaussian Components

摘要: This paper presents the development of a new numerical algorithm for statistical inference problems that require sampling from distributions which are intractable. We propose to develop our based on class Monte Carlo methods, Approximate Bayesian Computation (ABC), specifically designed deal with this type likelihood-free inference. ABC has become fundamental tool analysis complex models when likelihood function is computationally intractable or challenging mathematically specify. The central theme approach enhance current algorithms by exploiting structure mathematical via derivative information. introduce Progressive Correction Gaussian Components (PCGC) as efficient generating proposal in sampler. demonstrate two examples an acceptance rate one orders magnitude better than basic rejection sampling.