To effectively apply meta-learning to foundation models, a clear grasp of its core ideas is necessary. This chapter serves as a refresher on these foundational concepts, establishing a solid base for the advanced techniques that follow.

We begin by formally defining the meta-learning problem structure, detailing the roles of meta-training, meta-testing, tasks, support sets $S_{i}$, and query sets $Q_{i}$. Following this, we will examine a structured classification of meta-learning algorithms, differentiating between gradient-based, metric-based, and optimization-based methods.

A significant part of this chapter addresses the unique difficulties encountered when scaling meta-learning approaches to handle large foundation models, including computational cost, data requirements, and stability. Lastly, we will review standard benchmarks and established protocols essential for the rigorous evaluation of few-shot adaptation performance. This review ensures a common understanding before proceeding to more complex implementations and theoretical analyses.