Neural Tangent Kernel Theory Quiz

Answers and explanations appear as you take the quiz.

1. In the infinite-width limit, how does the Neural Tangent Kernel behave during training?
   • It stays essentially constant
   • It changes rapidly every step
   • It becomes zero
   • It grows exponentially
2. What is the NTK mathematically?
   • The Hessian of the loss
   • The inner product of the network's parameter gradients for two inputs
   • The softmax of the outputs
   • The variance of the weights
3. Who introduced the Neural Tangent Kernel?
   • Hinton and Salakhutdinov
   • LeCun and Bengio
   • Jacot, Gabriel, and Hongler (2018)
   • Vaswani and colleagues
4. What is the 'lazy training' regime?
   • When training is paused
   • When the optimizer uses no momentum
   • When data is fed in slowly
   • When parameters barely move from initialization and the network behaves linearly
5. A key limitation of NTK theory for real networks is that it misses:
   • Feature learning in finite-width networks
   • The role of gradient descent
   • The existence of weights
   • The use of training data
6. Because the NTK reduces training to kernel regression, what becomes possible?
   • Eliminating the need for any data
   • Predicting the trained output without fully training
   • Running models without parameters
   • Guaranteeing zero test error always
7. In the infinite-width limit, what probabilistic model are neural networks equivalent to at initialization?
   • A hidden Markov model
   • A decision tree ensemble
   • A Gaussian process
   • A finite-state game controller
8. For least-squares regression in the NTK limit, what mathematical form does the network function's training evolution follow?
   • A discrete rule that randomly changes labels
   • A decision tree that grows one branch per epoch
   • A nonlinear search over audio frames
   • A linear differential equation in function space