AI RESEARCH
HIFICL: High-Fidelity In-Context Learning for Multimodal Tasks
arXiv CS.CV
•
ArXi:2603.12760v1 Announce Type: new In-Context Learning (ICL) is a significant paradigm for Large Multimodal Models (LMMs), using a few in-context nstrations (ICDs) for new task adaptation. However, its performance is sensitive to nstration configurations and computationally expensive. Mathematically, the influence of these nstrations can be decomposed into a dynamic mixture of the standard attention output and the context values. Current approximation methods simplify this process by learning a "shift vector". Inspired by the exact decomposition, we