Question: Explain how the goal of domain adaptation research differs from the strategy recommended for improving one specific machine learning system.

Sample answer: Domain adaptation research focuses on building methods that train on one data distribution and still perform well on another, which makes it a broad and intellectually interesting problem. For a single application, it is usually more practical to create the development and test sets from the same distribution as the intended deployment data, because that gives clearer feedback and lets the team iterate faster.

Key points:

• Domain adaptation studies transfer across different data distributions.
• It is primarily a research problem.
• For a specific application, the development and test sets should usually match each other and the target use case.
• Matching the data distributions for evaluation is often more efficient than trying to solve transfer early on.

Rubric: The response should contrast the research goal of transferring across distributions with the practical advantage of using matching development and test sets for a specific application.