Nesterov Accelerated Gradient Quiz

Answers and explanations appear as you take the quiz.

1. What is the defining idea of Nesterov Accelerated Gradient compared to classical momentum?
   • It removes momentum entirely
   • It uses a fixed learning rate of 1.0
   • It evaluates the gradient at a look-ahead point rather than the current position
   • It only works without mini-batches
2. What provable convergence rate does Nesterov's method achieve on smooth convex problems?
   • Order 1/k^2
   • Order 1/k
   • Order 1/log k
   • No guaranteed rate
3. Who originally introduced this accelerated gradient method?
   • Geoffrey Hinton in 2012
   • Isaac Newton
   • John von Neumann
   • Yurii Nesterov in 1983
4. Why does the look-ahead gradient help near curved minima?
   • It increases the learning rate
   • It anticipates overshoot and applies a correction before it happens
   • It deletes the velocity term
   • It randomizes the step direction
5. In practice, how does the computational cost of Nesterov momentum compare to classical momentum?
   • Roughly the same, thanks to an algebraically rearranged update
   • About ten times higher
   • It requires second derivatives every step
   • It cannot be computed on GPUs
6. Which adaptive optimizer combines Nesterov look-ahead with Adam-style scaling?
   • RMSprop
   • Adagrad
   • Nadam
   • LBFGS
7. Using the stated look-ahead form, theta = 10, learning rate = 0.1, beta = 0.9 and velocity = 2 give which evaluation point?
   • 10.18
   • 9.82
   • 8.20
   • 10.90
8. If beta is zero in the guide’s Nesterov look-ahead formula, where is the gradient evaluated?
   • At a point displaced by the full old velocity
   • At the previous validation score
   • At an arbitrary future minibatch
   • At the current parameter position