You are correct that the papers I alluded to focus on finite n,

however, our analysis is not confined to “tail”—it’s for general k.

Also, asymptotic temporal behavior can readily be obtained by taking the limit of large n in the given solution.

In any case, I thought those papers were relevant and you might enjoy reading them, which is why I mentioned them. I wish you all the best for your future research in this direction.

Cheers.

]]>yes I have seen these papers: they study a different question in the sense that they look at the influence of the seed for FINITE n. This is a major difference in my opinion (for instance they consider the behavior of the tail of degrees, which is clearly unaffected by the seed for large n).

]]>Am I missing something here? Sorry if I misunderstood the content.

Cheers, Ziye

]]>Thanks for this blog article. This helps alot to understand Nesterov’s method. I have been exploring and reading about Nesterov’s convergence bound for quite some time now. I have a question: Are there some examples of simple and strongly convex functions for which Nesterov’s bound is better than the strongly convex bound and worse than the convex quadratic bound.

best,

Bikash

Could you show the use of FISTA in the case of the f(x) and g(x) you chose above (Namely g(x) is the l1 norm)?

Thank You.

]]>I have one question though: isn’t FTRL with the entropic barrier supposed to be equivalent with GeometricHedge of DHK08 (modulo discretization)? I thought this was the case but you don’t mention the connection here. ]]>