I didn't completely understand the SMO algorithm's procedure.

at the first iteration we assume a2,…,am are fixed and optimize the problem according to a1, I assume by finding the derivative of the function according to a1.

whats the next step? from the derivative you can get the optimal value for a1, which is **dependent** of a2,…,am which are assumed to be fixed but are actually still variables.

do you then plug the optimal a1 into the original problem and optimize the new problem according to a2 an so on?

from what I understood in class, after optimizing the problem according to some ai, we get a optimal value for ai which isn't dependent of the other aj (an actual number).

specifically in this exercise, do we need to optimize the problem according to a1 only?

and in subsection d, what do you mean exactly by the computational complexity? of calculating the derivative? since you can't really do anything with the fixed variables.

thanks!