Bayesian Estimation of Big Data and Big Models

Lead CI: Robert Kohn

The first purpose of the research is to develop Bayesian or quasi Bayesian methods for handling Big Data and BigModels data using a number of approaches.

Our first approach will be fully Bayesian and involves randomly subsampling the date at each iteration using MCMC. Our second approach is also fully Bayesian and is based on simulated annealing. The third approach will involve Approximate Bayesian Computation. Our fourth approach is based on variational Bayes methods. Our fifth approach is approximate and based on stochastic approximation methods. The second purpose of the research is to apply the methodology to a number of challenging data sets, and in particular to apply it to panel data consisting of a large number of individuals playing games on apps, where we have a large number of observations for each individual.