Project B2 - Crossmodal inference by conjoining probabilistic and symbolic models

PIs: Prof. Dr. Jun Zhu, Prof. Dr. Jan Gläscher

Project B2 will continue the work from phase 1 on visual-text integration using Bayesian methods, in particular Regularized Bayes, which was the focus during Phase 1. However, during this funding phase, probabilistic deep learning networks will be contrasted and combined with symbolic networks in their efficiency and success in visual image and text integration. The former approach has advantages in modelling uncertainty and probabilistic inference, while the latter can capture the relationships between multimodal objects that are semantically interpretable. Joining these two approaches will result in interpretable crossmodal inference and prediction of visual-text data.