In-Context Guidance: Learning Inter-Task Synergies via Numerical Foundational Models for Few-Shot Multitask Optimization

This paper presents ICG-MTO, a novel framework for multi-task optimization that employs numerical foundational models to enhance inter-task relationship estimation under limited evaluation budgets. The approach specifically targets the challenge of negative transfer in few-shot scenarios, where traditional methods struggle to accurately capture task dependencies. By leveraging foundation model capabilities, the framework aims to improve optimization performance across related tasks without extensive per-task evaluations.