Why might an end-to-end system underperform a hand-engineered pipeline when training data is scarce?

In a very small-data setting, most algorithmic knowledge may need to come from human insight.

With very little training data, most knowledge may need to come from human _____.

Match each small-data concept with its source-grounded meaning.

Order the reasoning for choosing hand engineering in a small-data setting.

Explain why small training sets can increase the value of hand-engineered components.

Decide how to supply knowledge when an end-to-end system has very little training data.

What supplies most of an algorithm's knowledge when its training set is very small?

Which design response best follows from having a very small training set?

A small training set guarantees that every hand-engineered pipeline will beat every end-to-end system.