Meta-Learning
Training a shared starting point across many related tasks, rather than a solution to one, so adapting to any new task takes only a handful of gradient steps.
MAML's outer loop evaluates parameters after a simulated inner-loop gradient step, directly optimizing for how easy a starting point is to adapt from rather than how good it is on its own. Prototypical networks skip gradient-based adaptation entirely, learning an embedding space where averaging a few labeled examples into a prototype is enough to classify.
A meta-learned starting point only helps for new tasks that resemble the distribution it was trained across.