Disentangled Representation Learning via Modular Compositional Bias | Read Paper on Bytez