Hierarchical Monte Carlo Fusion
Monte Carlo Fusion [Dai, Pollock & Roberts 2018] proposes a new theory and methodology to tackle the problem of unifying distributed analyses and inferences on shared parameters from multiple sources, into a single coherent inference. This problem can appear in settings such as expert elicitation, distributed ‘big data’ problems, and tempering. However, the original Monte Carlo fusion algorithm is inefficient in some settings, for instance when the number of sub-posteriors to combine is large. Here, we introduce ‘Hierarchical Monte Carlo Fusion’ which proposes a new framework to perform fusion with the aim to alleviate this problem.
Engineered world Theoretical foundations