Variational Structured Attention Networks for Deep Visual Representation Learning | Read Paper on Bytez