Transfer Learning

A technique where knowledge gained from training on one task is applied to a different but related task, rather than starting from scratch. Transfer learning dramatically reduces the data and compute needed for new tasks and is the basis for the dominant paradigm in modern AI, pre-train a large model on broad data, then adapt it to specific applications.