Conference paperContext-sensitive prediction of facial expressivity using multimodal hierarchical Bayesian neural networks