0
$\begingroup$

I am working on the development of novel optimization algorithm where it DOES NOT suffer from the non-differentiability (cf. subgradient), which also works well in differentiable class of the objectives.

Now, I want to explore the behaviour of my algorithm. I only took relavant optimization courses like 4 lectures, but it does not necessarily mean I am super good or much experienced in this field.

So, is there any good article which $\mathbf{LISTS}$ interesting known convex objectives at a single sight for this purpose? Or should I manually try this and that based on my own experience so far?

$\endgroup$
  • 1
    $\begingroup$ You can search for "review article". Those are articles broadly speaking trying to summarize the current state of a topic. Well, if it is new anyway. $\endgroup$ – mathreadler May 7 '18 at 7:00
  • 1
    $\begingroup$ The search term you need is benchmark problems. $\endgroup$ – Rahul May 7 '18 at 7:21
  • $\begingroup$ Such a method already exists. See optimization-online.org/DB_HTML/2013/04/3833.html. Also, a convex function on an open set is twice differentiable almost everywhere, so there is no practical difference between differentiability and subdifferentiability for convex problems. $\endgroup$ – Serg May 7 '18 at 7:38
  • $\begingroup$ @Serg , "There is no practical difference between differentiability and subdifferentiability for convex problems". It is hard to say so. I shortly checked your reference, clearly it is sublinear, which is a typical story. My goal is to achieve even superlinear. $\endgroup$ – Robin May 8 '18 at 7:42
  • $\begingroup$ @Robin. The method i linked to is optimal. This is as good as it gets without additional assumptions like strong convexity. Even linear global convergence is impossible to achieve for non strongl convex objectives. $\endgroup$ – Serg May 8 '18 at 7:54

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.