What is correct about stochastic gradient descent? (select all that apply)1 pointThe loss may exhibit sudden increases The loss must be linearIt's an approximation of batch gradient descent

The correct statements about stochastic gradient descent are:

1. The loss may exhibit sudden increases: This is true because stochastic gradient descent updates the model's parameters for each training example one by one, leading to frequent and sometimes drastic changes in the loss function.

2. It's an approximation of batch gradient descent: This is also true. While batch gradient descent computes the gradient using the whole dataset, stochastic gradient descent approximates this by computing the gradient one example at a time.

The statement "The loss must be linear" is not correct. The loss function in stochastic gradient descent does not need to be linear. It can be any differentiable function.