Question: Write a concise analytical response connecting small training sets, expert knowledge, hand-designed components, and possible model performance.

Sample answer: When the training set is very small, a fully learned system may not have enough examples to discover all of the useful structure on its own. A hand-designed pipeline can inject knowledge from human experts into the system. Because that expert knowledge is already built into the pipeline, it may outperform a purely learned end-to-end model in a low-data setting. For that reason, small datasets often increase the value of hand-designed components as a way to supply algorithmic knowledge.

Key points:

• A very small training set limits what can be learned from data
• Human experts contribute useful algorithmic knowledge
• Hand-designed components encode that knowledge
• A fully learned system may perform worse in the low-data case

Rubric: A strong response should explain the small-data limitation, identify expert knowledge as the source of much of the algorithm's knowledge, connect that knowledge to hand-designed components, and state that the end-to-end system may perform worse rather than always doing so.