This paper introduces a novel Multimodal Attention-based Normalizing Flow approach to developing explicit, interpretable, and tractable multimodal fusion learning