Three positions are available in the following project of PI’s Prof. Aki Vehtari (Aalto University), Prof. Antti Honkela and Prof. Arto. Klami (University of Helsinki).
We will develop theory and methods for assessing the quality of distributional approximations based on leave-one-out cross-validation and projection predictive model reductions, and seek to improve the inference accuracy by targeting the approximation towards the eventual application goal and by better utilising the available data, e.g., when having data with privacy constraints.