Definition
A machine learning technique where a model trained for one task is reused as the starting point for a model on a different, but related, downstream task.
In short: A machine learning technique where a model trained for one task is reused as the starting point for a model on a different, but related, downstream task.