Technical report detail

Efficient and Portable MPI Support for Approximate Bayesian Computation

by Lorenzo Fabbri, Avinash Ummadisingu


In order to obtain a mechanistic understanding of a natural or social system, it is sometimes useful to employ complex stochastic models. These models, on the other hand, lead almost inevitably to computationally intractable likelihood functions. Since the posterior distribution of the model s parameters is proportional to the likelihood function, any problem with the evaluation of the former leads to difficulties in the process of learning these parameters. For this reason, many methods, the so-called likelihood-free methods, have gained popularity in the literature. The contribution of this project is the development of a parallel backend for ABCpy based on MPI. The analysis of scaling properties on the CSCS supercomputer Piz Daint is also provided.


Technical report 2017/04, June 2017

BibTex entry

@techreport{17efficient, author = {Lorenzo Fabbri and Avinash Ummadisingu}, title = {Efficient and Portable MPI Support for Approximate Bayesian Computation}, institution = {University of Lugano}, number = {2017/04}, year = 2017, month = jun }
Attachments