What is transfer learning?

Correct answer: Using a pre-trained model as a starting point for a new task

Option B is correct because transfer learning involves taking a model that was already trained on a large dataset (often a general task like image classification on ImageNet or language modeling) and fine-tuning or adapting it for a different but related downstream task, saving significant compute and data requirements. This technique is fundamental to modern ML practice, powering models like BERT, GPT, and ResNet variants used in production. Option A is incorrect because splitting training across multiple machines describes distributed or parallel training, not transfer learning. Option C is incorrect because converting a model between frameworks (e.g., PyTorch to TensorFlow) is model serialization or interoperability, not transfer learning. Option D is incorrect because moving a trained model between cloud providers is a deployment and portability concern, entirely unrelated to the learning paradigm.