Approximate Bayesian computation

Many modern statistical settings feature the analysis of data that may arise from unknown generating processes, or processes for which the generative models are computationally infeasible to interact with. Conventional estimation and inference solution methods in such settings may be unwieldy or impossible to implement. The approximate Bayesian computation (ABC) approach is a potent method in such scenarios, since it does not require the knowledge of the underlying generative model in order to perform inference. Furthermore, when combined with sufficiently regular discrepancy measurements such as the energy statistic, ABC can be shown to have desirable asymptotic properties. We provide a concise introduction to the general ABC framework.Survey: https://www.surveymonkey.com/r/T9PKDPT

诸多现代统计学研究场景中，研究者需要分析两类数据：一类源自未知生成过程，另一类对应的生成模型在计算层面难以实现交互。在此类场景中，常规的估计与推断求解方法往往难以操作，甚至无法落地实施。近似贝叶斯计算（approximate Bayesian computation, ABC）方法正是此类场景下的有力工具，因其无需掌握底层生成模型即可完成统计推断任务。进一步而言，当与能量统计量这类具备良好正则性的差异度量结合使用时，ABC方法可被证明拥有优良的渐近特性。本文将对通用ABC框架进行简要介绍。调研问卷：https://www.surveymonkey.com/r/T9PKDPT