Week 2: Training And Gradients

Objective

Understand how a model improves by measuring loss, computing gradients, and updating parameters with gradient descent.

Required Videos

• 3Blue1Brown: “Gradient descent, how neural networks learn”
• Optional: any short derivative refresher if slopes feel fuzzy

Tasks

1. Run python week2_training.py.
2. Open the saved loss plot and describe its shape.
3. Change the learning rate and compare the results.
4. Change the number of training steps and see what happens.

Deliverables

• A saved loss plot
• A note describing which learning rate worked best
• A short explanation of why the loss changed over time

Checkpoint Questions

• What does the loss function measure?
• What are we changing after each training step?
• Why is the gradient useful?
• What can go wrong if the learning rate is too high?

Stretch Idea

Add random noise to the training data and see how the final fit changes.