Consider a 2-dimensional weight space, and two different error

functions:

E([w,w]) = w + 4 w - 97 w + 13 w

E([w,w]) = 1000 w + 10 w + 7 w - 3 w

If you optimize each of these using batch gradient descent, with the

learning rate set as high as you can without the system oscillating,

what is the highest learning rate you can use for each of these? What is

each of their rates of convergence? Which is faster? Please show the

relevant formula and your calculations, and draw a diagram.