Week 3: Micrograd

Objective

Study a tiny autodiff engine closely enough that backprop stops feeling like magic.

Required Videos

• Andrej Karpathy: “The spelled-out intro to neural networks and backpropagation”
• Optional: revisit Week 2 before watching if gradients still feel abstract

Tasks

1. Read micrograd_notes.md.
2. Watch the required Karpathy video with a notebook open.
3. Complete the guided exercise in exercise.md.
4. Draw one small computation graph by hand and annotate the gradient flow.

Deliverables

• Annotated notes on the Value object
• A completed exercise with your own comments
• One hand-drawn or typed computation graph

Checkpoint Questions

• Why does each Value need to know its parents?
• Why do gradients add together from multiple paths?
• What happens if backward passes happen in the wrong order?
• What did micrograd make clearer than Week 2?

Suggested Workflow

Do not rush. Pause the video whenever the graph grows and trace the local derivative yourself.