Interpretable probabilistic embeddings: bridging the gap between topic models and neural networks | Read Paper on Bytez