New method reveals how vision and language models actually work together

Original: Towards Understanding Modality Interaction in Multimodal Language Models via Partial Information Decomposition

Source: arxiv.org ↗