#mini-batch

Question 1: Select all that apply

Which statements about full-batch gradient descent, SGD and mini-batch gradient descent are correct? Select all that apply.

When the examples in a mini-batch are drawn uniformly at random, the mini-batch gradient is an unbiased estimate of the full-batch gradient1
Averaging the gradient over a larger mini-batch reduces the variance of this estimate2
On a non-convex loss, full-batch gradient descent reaches the global minimum because its gradient is exact3
The three methods use different update rules, which is why they reach different solutions4