Calculus for ML Quiz

Calculus for ML Quiz

Work through these questions to check your understanding of the calculus tools used throughout ML. Each question covers one of the core ideas from this course.

Exercise

f(x, y) = x²y³. What is ∂f/∂y?

Need a hint?

Treat x as a constant.

Exercise

Minimize f(w) = w² by gradient descent from w = 2 with learning rate η = 0.25. Where is w after one step?

Need a hint?

w ← w − η·f′(w), and f′(w) = 2w.

Exercise

What is d/dx of sin(x²)?

Need a hint?

Outer derivative evaluated at the inner value, times the inner derivative.

Exercise

At a critical point, the Hessian has eigenvalues 2 and −3. What kind of critical point is it?

Need a hint?

Mixed signs mean the surface curves up in one direction and down in another.

Exercise

f maps R⁴ → R². What is the shape of its Jacobian?

Need a hint?

Rows index outputs, columns index inputs.

Exercise

A network has a million parameters and one scalar loss. Roughly what does computing ALL the loss gradients via backpropagation cost?

Need a hint?

Backprop is one vector-Jacobian product per layer, reusing cached activations.

Exercise

In the ML project loop, backpropagation (the chain rule on a computational graph) primarily serves which slot?

Need a hint?

Backprop produces the gradients that a parameter-search algorithm consumes.

Exercise

Transfer test — a technique this course did NOT teach: batch normalization inserts a normalization step between layers so training converges faster and more stably. Which slot does it primarily modify?

Need a hint?

It reconditions the loss surface so gradient descent moves better — it doesn't add a new function family or change the loss.