How does Gradient Descent help in optimizing Multiple Linear Regression