Tag
Proposes ROAM, a framework that uses LLM world knowledge and reasoning to adapt frozen specialist models to unseen scenarios without retraining, achieving over 20% MAE reduction with minimal overhead.