Correlation Information Bottleneck: Towards Adapting Pretrained Multimodal Models for Robust Visual Question Answering | Read Paper on Bytez