What is meta-learning in cost function optimization?

Meta-learning optimizes learning algorithms themselves rather than a single cost function.

Meta-learning introduces a higher-level optimization problem where the objective is to learn how to learn. The cost function becomes nested: an outer loop optimizes model initialization or optimizer parameters, while an inner loop trains the model. This creates bilevel optimization problems.