Scaling Posterior Distributions over Differently-Curated Datasets: A Bayesian-Neural-Networks Methodology

This paper provides an introduction to an innovative methodology for scaling posterior distributions over differently-curated datasets. The proposed methodology is based on Bayesian Neural Networks, improved by effective sampling algorithms. These algorithms finally realize a suitable model setup for improving the scaling effect. Theoretical results are presented and discussed in details, as well as a modern case study focused on stock quotation prediction that confirms the successful application of our proposed methodology to emerging big data analytics settings.