HW4 1.3 Question about convex optimazation

[Yitong Zhou]

I think what I should do is minimizing the E(\phi(x))-empirical average, but since the Phi(x) is a two dimension function, and fminsearch of Matlab can only handle single-valued function, I applied a norm 2 of Phi(x), is it correct to do so? The experiments results look fine, though with some errors.

For another question, I find that the fminsearch can only find local optimal solution, is it ok to just start \theta from (0,0) to search? I feel it could be really hard to test out other local solutions.