Every compartment and flow of our life control from Bayes wisdom

[hyunjimoon]

From Aki's topic below, I think we can extract some wisdom in our life control; with emphasis on both "our (compartment)" and "control (flow)". Prior to meeting SD, I liked my work because people say its result is needed. Now I love every flow and compartment of my life; and their extremely high autocorrelation. Sincere thanks for resetting me to this difficult posterior state + for being correlated with me.

Topic: Bayesian workflows for iterative model building and networks of models 

Statistical analysis is critical when it comes to obtaining insights from data. Despite the practical success of iterative Bayesian statistical model building, it has been criticized to violate pure Bayesian theory and that we may end up with a different model had the data come out differently. In this project, we formalize and develop theory and diagnostics for iterative Bayesian model building. The practical workflow recommendations and diagnostics guide the modeller through the appropriate steps to ensure safe iterative model building, or indicate when the modeler is likely to be in the danger zone. Sometimes the user wants also to build and compare models of different complexity or based on different assumptions.  Similar workflow ideas can be used to support also analysis of networks of models, making it easier to illustrate similarities and important differences. By making applied scientific research and data analysis more reliable and reproducible, our understanding of the world and decision-making will be improved. 

Topic: Evaluating and improving posterior inference for difficult posteriors

Both MCMC and distributional approximations (variational and Laplace approximations) often struggle to handle complex posteriors, but we lack good tools for understanding how and why. We study diagnostics for identifying the nature of the computational difficulty, e.g. whether the difficulty is caused by narrow funnels or strong curvature. We also develop improved inference algorithms, e.g. via automated and semi-automated transformations. Keywords: MCMC, variational approximation, differential geometry, inference diagnostics, Bayesian workflow 

Topic: Workflows for better priors

Bayesian models rely on prior distributions that encode knowledge about the problem, but specifying good priors is often difficult in practice. We are working on multiple fronts on making it easier, with contributions to e.g. prior elicitation, prior diagnostics, prior checking, and specification of priors in predictive spaces. Keywords: Prior elicitation, Bayesian workflow, priors on predictive space, default priors