Destilação e compressão de modelos bayesiano

This research project has as main goal the development of methods for compressing bayesian models. More specifically, we intend to distill intractable bayesian models using simplified substitute models. This work aims for expanding  results related to prior swapping via importance sampling, by optimizing the choice of substitute prior distributions. We expect this project provides a method for decreasing the computacional power needed in the training of complex  bayesian models, contributing to the Sustainable Development Goal 13 – action against the global climate change.