Week 13 Part IV · Integration

Integration & Transfer Learning

Transfer learning and fine-tuning; model inference; the end-to-end workflow into the advanced courses.

Learning goals

Fine-tune a pretrained model end-to-end.
Run inference and assemble a full workflow.
Reason about when transfer learning helps.

This is the weekly homework lab, completed independently after the lecture and the practice lesson. It follows the course's Build / Predict & probe / Explain & defend model: use an AI assistant freely for the Build; the graded learning is in Predict and Explain. See the AI-use policy and a fully worked sample submission.

⚙Exercise

Part A · AI assistant welcomeBuild

Fine-tune a pretrained model end-to-end on a new task and run inference.

Part B · student reasoningPredict & probe

Predict the ranking of from-scratch, fine-tuning, and frozen-features under a fixed budget.

Part C · in plain languageExplain & defend

Compare the three regimes, explain when transfer learning helps and why, and reflect on the full workflow.

✓Deliverables

A fine-tuning notebook with a three-regime comparison.
A reflection that doubles as final-project preparation. The final project is due this week.

Hints.

Use a smaller learning rate for pretrained layers; unfreeze gradually.
Match input preprocessing to the pretrained model.

❓Self-check

Answer each before expanding it. If one is unclear, revisit the lab and the references.

Name the two transfer-learning strategies.

Fixed feature extraction (freezing) and fine-tuning.

When does freezing the backbone make most sense?

When the target dataset is small and similar to the source.

Why use a smaller learning rate for pretrained layers?

To avoid destroying useful pretrained features.

Why match the pretrained model's preprocessing?

Inputs must match the normalization/distribution it was trained on.

What is the end-to-end workflow?

Data, model, train, evaluate, infer/deploy.

Instructor lesson plan (with references)