Introduction to first edition:

The first edition architecture followed the implementation flow chart presented by Rubin et al. [2010].  The flow chart therein detailed three modules - one for each of the core components of Bayesian analysis.  The software was developed to aid with problem formulation and prior probability distributions in Block 1, to calculate likelihood function estimates in Block 2, and to determine posterior probability distributions of model parameters in Block 3.  Many of the underlying components of the first edition of the software have been retained by the latest edition.  However, the original architecture has been revised for two main reasons.  First, the generality and flexibility of the interfaces needed to be increased, which will help the MAD software support Bayesian model inversion in different scientific areas in the future.  Second, the implementation is now a single module that handles problem formulation, Bayesian analysis, and all computational configuration steps, which means that the MAD software fully automates model inversion and will permit future parallelization of the process.

Block 1 Material:

Block 2/3 Material:

Functionality Diagram for Blocks I,II, and III



Meeting Notes:

Toy Problem 2:

Mad Interface:

Last edited Nov 8, 2013 at 9:19 PM by matthewoverUCB, version 7


No comments yet.