AdvancedMultiple choice
You increase the batch size from to without changing anything else. What happens?
- AGradient noise falls fourfold, so the learning rate usually has to rise to compensate
- BTraining becomes strictly better, since the gradient is more accurate
- CNothing changes, since the expected gradient is the same
- DConvergence slows in wall-clock time on any hardware
The worked solution is in Premium
The answer, the full working and the one idea to take away – for this and all 1,322 questions in the bank. Answer it in practice and your working is marked, with a known mistake named when you make one.
Learn the method
Reported in interviews at
More machine learning questions
- One step of gradient descent, part 1 of 4Foundation
- Gradient descent with momentum uses a learning rate of 0.01 and β = 0.9.Applied
- Gradient descent on f(x) = x² starts at x = 1 with learning rate 0.1.Applied
- A quadratic loss has Hessian eigenvalues 4 and 0.5.Applied
- One step of gradient descent, part 2 of 4Applied
- Adam’s first-moment estimate starts at 0 with β₁ = 0.9.Advanced