Skip to main content

Timeline for On a version of gradient descent

Current License: CC BY-SA 3.0

11 events
when toggle format what by license comment
May 17, 2013 at 18:30 vote accept user21162
May 17, 2013 at 18:30 comment added user21162 Gotcha! I should learn to be more handy with code myself...
May 17, 2013 at 17:38 comment added Suvrit I tested it on a simple numerical example; the sequence $r_t$ is not monotonic. Since you essentially brought up this point, I attributed it to you.
May 17, 2013 at 17:11 comment added user21162 ...you say I observed that $r_t$ need not be monotonically decreasing, but I'm still uncertain about this: in my example above, $\langle f_i'(x_t) e_i, x_t - x^{\rm min} \rangle < 0$ only for the $i$ such that $|f_i'(x_t)|$ is not maximal..
May 17, 2013 at 17:09 comment added user21162 Thank you! Ok, I understand: so the answer to my question is that my claim ``the author appears to be using...'' is not right.
May 17, 2013 at 16:43 history edited Suvrit CC BY-SA 3.0
fixed answer
May 17, 2013 at 6:49 comment added Suvrit Ok, I think I was too quick!! I'll check the details of the proof tmw and update my answer.
May 17, 2013 at 6:28 comment added user21162 Here is what seems to me to be a counterexample to the inequality in my comment which you say is true: Suppose $$f(x,y) = 10*(x-y-1)^2+(x+y+1)^2$$ Clearly $f$ is convex and moreover the minimum is $$(x,y)^{\rm min} = (0,-1)$$ Now computing the derivative: $$f_{x}(x,y) = 20*(x-y-1) + 2 (x+y+1)$$ $$f_x(0.5,0) = 20*(-.5) + 2(1.5)=-7$$ so that $$ \langle f_x(0.5,0) e_1, (0.5,0)-(0,-1) \rangle = -7*0.5 < 0$$
May 17, 2013 at 5:45 comment added Suvrit Sorry, don't have time to type it in right now; please see (2.1.7 and 2.1.8) in the book Introductory lectures on convex optimization by Yu. Nesterov; you'll have to adapt those lemmas to the case of $f$ restricted to the $i$-th coordinate. From that the desired conclusion will follow. (Also, since $f$ is convex, $f'$ is monotone, so the inequality that you've mentioned in the comment also holds. The index being largest in magnitude was used for an earlier inequality in the cited paper.
May 16, 2013 at 20:30 comment added user21162 Yes, please fill in the details. It is not in general true that $\langle f_i'(x_t) e_i, x_t - x^* \rangle \geq 0$, is it? I think if that is true, its only true for the $i$ such that $|f_i'(x_t)|$ is largest.
May 16, 2013 at 19:20 history answered Suvrit CC BY-SA 3.0