Multimodal In-Context Learning with Explicit Rule Induction.

arXiv:2610.41754v1 Announce Type: new Abstract: A central challenge in multimodal in-context learning (MICL) arises from tasks that are ill-defined without context and require uncovering the underlying rules from demonstrations. Existing methods predominantly