Using knowledge gained from training on one task to improve performance on a different but related task. Like a chef who learned French cuisine using those knife skills to quickly learn Japanese cooking.
A model trained to recognize cats and dogs can be fine-tuned with transfer learning to identify specific dog breeds, requiring far less training data.
All four platforms support transfer learning by letting you start from a pre-trained model (often from a model hub), then fine-tune it on your own labeled data using managed training jobs, GPUs, and deployment endpoints.