There is bit of difference between fitting dataset to some convenient parameterized function vs finding global minima of non-convex function What's the differe

There is bit of difference between fitting dataset to some convenient parameterized function vs finding global minima of non-convex function What's the difference? Any point where the loss is zero is global minimum.