Aligning Multimodal Representations through an Information Bottleneck | Read Paper on Bytez