few-shot

A machine learning paradigm where a model is trained on a very small amount of data, typically less than that required for traditional machine learning approaches.