Current browse context:
math.OC
Change to browse by:
References & Citations
Mathematics > Optimization and Control
Title: New nonasymptotic convergence rates of stochastic proximal pointalgorithm for convex optimization problems
(Submitted on 22 Jan 2019 (v1), last revised 2 May 2020 (this version, v4))
Abstract: Large sectors of the recent optimization literature focused in the last decade on the development of optimal stochastic first order schemes for constrained convex models under progressively relaxed assumptions. Stochastic proximal point is an iterative scheme born from the adaptation of proximal point algorithm to noisy stochastic optimization, with a resulting iteration related to stochastic alternating projections. Inspired by the scalability of alternating projection methods, we start from the (linear) regularity assumption, typically used in convex feasiblity problems to guarantee the linear convergence of stochastic alternating projection methods, and analyze a general weak linear regularity condition which facilitates convergence rate boosts in stochastic proximal point schemes. Our applications include many non-strongly convex functions classes often used in machine learning and statistics. Moreover, under weak linear regularity assumption we guarantee $\mathcal{O}\left(\frac{1}{k}\right)$ convergence rate for SPP, in terms of the distance to the optimal set, using only projections onto a simple component set. Linear convergence is obtained for interpolation setting, when the optimal set of the expected cost is included into the optimal sets of each functional component.
Submission history
From: Andrei Patrascu [view email][v1] Tue, 22 Jan 2019 15:10:59 GMT (22kb)
[v2] Sat, 9 Feb 2019 08:57:26 GMT (23kb)
[v3] Wed, 7 Aug 2019 12:24:13 GMT (32kb)
[v4] Sat, 2 May 2020 09:43:30 GMT (328kb,D)
Link back to: arXiv, form interface, contact.