# How can stochastic gradient descent avoid the problem of a local minimum?

I know that stochastic gradient descent has random behavior, but I don't know why.

-