Login to unlock
Deep Learning · Phase 7: Attention & Transformers · Pretraining & Transfer Learning

Pretraining & Transfer Learning — Checkpoint Quiz

Checkpoint quiz 3 questions
Back to course

Check your understanding of this module before moving on.

?

Checkpoint quiz 3 questions

A quick check on this module's key ideas before you move on
0/3
Question 01

What is the core idea behind transfer learning in deep learning?

A is correct. Transfer learning leverages the general features (edges, textures, syntax, semantics, etc.) a model already learned on a large dataset, reusing that knowledge as a strong starting point for a new task rather than relearning everything from zero.
Question 02

Why does fine-tuning a pretrained model often require far less labeled data than training from scratch?

A is correct. Much of what a model needs to learn (general low-level and mid-level features) has already been captured during pretraining on a large dataset. Fine-tuning only has to adjust or specialize those features, which takes far less task-specific data than learning from nothing.
Question 03

What is a common strategy when fine-tuning a pretrained network on a small new dataset, to avoid destroying its useful pretrained features?

A is correct. Freezing earlier layers preserves the general features they've already learned, while training only the later, more task-specific layers (often with a small learning rate) adapts the model without catastrophically overwriting the useful pretrained representations.