Transfer Learning

Instead of training a new model entirely from scratch, teams start from a pre-trained model and adapt it using additional data or task-specific training. Transfer learning reduces data and compute requirements and is foundational to many modern vision and language applications.