demonstration-retrieval

This paper introduces DOPA, a demonstration search framework that uses an out-of-distribution proxy to retrieve robust demonstrations for LLMs when the target domain is inaccessible, enhancing in-context learning performance under distribution shift.