ResearchTrend.AI
  • Papers
  • Communities
  • Events
  • Blog
  • Pricing
Papers
Communities
Social Events
Terms and Conditions
Pricing
Parameter LabParameter LabTwitterGitHubLinkedInBlueskyYoutube

© 2025 ResearchTrend.AI, All rights reserved.

  1. Home
  2. Papers
  3. 1501.02414
26
11

Convergence Rates and Decoupling in Linear Stochastic Approximation Algorithms

11 January 2015
M. Kouritzin
Samira Sadeghi
ArXiv (abs)PDFHTML
Abstract

Almost sure convergence rates for linear algorithms hk+1=hk+1kχ(bk−Akhk)h_{k+1} = h_k +\frac{1}{k^\chi} (b_k-A_kh_k)hk+1​=hk​+kχ1​(bk​−Ak​hk​) are studied, where χ∈(0,1)\chi\in(0,1)χ∈(0,1), {Ak}k=1∞\{A_{k}\}_{k=1}^\infty{Ak​}k=1∞​ are symmetric, positive semidefinite random matrices and {bk}k=1∞\{b_{k}\}_{k=1}^\infty{bk​}k=1∞​ are random vectors. It is shown that ∣hn−A−1b∣=o(n−γ)|h_n- A^{-1}b|=o(n^{-\gamma})∣hn​−A−1b∣=o(n−γ) a.s. for the γ∈[0,χ)\gamma\in[0,\chi)γ∈[0,χ), positive definite AAA and vector bbb such that 1nχ−γ∑k=1n(Ak−A)→0\frac{1}{n^{\chi-\gamma}}\sum\limits_{k=1}^n (A_{k}- A)\to 0nχ−γ1​k=1∑n​(Ak​−A)→0 and 1nχ−γ∑k=1n(bk−b)→0\frac{1}{n^{\chi-\gamma}}\sum\limits_{k=1}^n (b_k-b)\to 0nχ−γ1​k=1∑n​(bk​−b)→0 a.s. When χ−γ∈(12,1)\chi-\gamma\in\left(\frac12,1\right)χ−γ∈(21​,1), these assumptions are implied by the Marcinkiewicz strong law of large numbers, which allows the {Ak}\{A_k\}{Ak​} and {bk}\{b_k\}{bk​} to have heavy-tails, long-range dependence or both. Finally, corroborating experimental outcomes and decreasing-gain design considerations are provided.

View on arXiv
Comments on this paper