Microsoft Azure AI Engineer Associate AI-102 Practice Question

The neural build mode is designed to generalize across many layouts by using transfer-learning techniques. It can be trained with as few as five labeled documents to start and allows multiple layouts in a single model, so you avoid creating and labeling separate models for every supplier. Template mode relies on a consistent visual template and typically needs at least five labeled examples per layout, making it impractical here. A layout model returns only structural text information, not mapped key-value fields. A composed model first requires training individual extraction models (template or neural) and therefore does not reduce the labeling effort in this scenario.