We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:

Download:

Current browse context:

cs.LG

Change to browse by:

References & Citations

DBLP - CS Bibliography

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Computer Science > Machine Learning

Title: Adversarial Delays in Online Strongly-Convex Optimization

Abstract: We consider the problem of strongly-convex online optimization in presence of adversarial delays; in a T-iteration online game, the feedback of the player's query at time t is arbitrarily delayed by an adversary for d_t rounds and delivered before the game ends, at iteration t+d_t-1. Specifically for \algo{online-gradient-descent} algorithm we show it has a simple regret bound of \Oh{\sum_{t=1}^T \log (1+ \frac{d_t}{t})}. This gives a clear and simple bound without resorting any distributional and limiting assumptions on the delays. We further show how this result encompasses and generalizes several of the existing known results in the literature. Specifically it matches the celebrated logarithmic regret \Oh{\log T} when there are no delays (i.e. d_t = 1) and regret bound of \Oh{\tau \log T} for constant delays d_t = \tau.
Comments: We discovered mistakes in the proof of proof of Theorem 3.1. The overall is no longer correct, although the claim is still true
Subjects: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Machine Learning (stat.ML)
Cite as: arXiv:1605.06201 [cs.LG]
  (or arXiv:1605.06201v4 [cs.LG] for this version)

Submission history

From: Daniel Khashabi Mr. [view email]
[v1] Fri, 20 May 2016 02:55:59 GMT (86kb,D)
[v2] Wed, 22 Feb 2017 18:41:57 GMT (0kb,I)
[v3] Mon, 9 Sep 2019 18:34:56 GMT (5016kb,D)
[v4] Wed, 11 Sep 2019 04:53:49 GMT (0kb,I)

Link back to: arXiv, form interface, contact.