English

Parallel algorithms for maximizing one-sided $\sigma$-smooth function

Optimization and Control 2022-06-14 v1

Abstract

In this paper, we study the problem of maximizing a monotone normalized one-sided σ\sigma-smooth (OSSOSS for short) function F(x)F(x), subject to a convex polytope. This problem was first introduced by Mehrdad et al. \cite{GSS2021} to characterize the multilinear extension of some set functions. Different with the serial algorithm with name Jump-Start Continuous Greedy Algorithm by Mehrdad et al. \cite{GSS2021}, we propose Jump-Start Parallel Greedy (JSPG for short) algorithm, the first parallel algorithm, for this problem. The approximation ratio of JSPG algorithm is proved to be ((1e(αα+1)2σ)ϵ)((1-e^{-\left(\frac{\alpha}{\alpha+1}\right)^{2\sigma}}) \epsilon) for any any number α(0,1]\alpha\in(0,1] and ϵ>0\epsilon>0. We also prove that our JSPG algorithm runs in (O(logn/ϵ2))(O(\log n/\epsilon^{2})) adaptive rounds and consumes O(nlogn/ϵ2)O(n \log n/\epsilon^{2}) queries. In addition, we study the stochastic version of maximizing monotone normalized OSSOSS function, in which the objective function F(x)F(x) is defined as F(x)=EyTf(x,y)F(x)=\mathbb{E}_{y\sim T}f(x,y). Here ff is a stochastic function with respect to the random variable YY, and yy is the realization of YY drawn from a probability distribution TT. For this stochastic version, we design Stochastic Parallel-Greedy (SPG) algorithm, which achieves a result of F(x)(1e(αα+1)2σϵ)OPTO(κ1/2)F(x)\geq(1 -e^{-\left(\frac{\alpha}{\alpha+1}\right)^{2\sigma}}-\epsilon)OPT-O(\kappa^{1/2}), with the same time complexity of JSPG algorithm. Here κ=max{5F(x0)do2,16σ2+2L2D2}(t+9)2/3\kappa=\frac{\max \{5\|\nabla F(x_{0})-d_{o}\|^{2}, 16\sigma^{2}+2L^{2}D^{2}\}}{(t+9)^{2/3}} is related to the preset parameters σ,L,D\sigma, L, D and time tt.

Keywords

Cite

@article{arxiv.2206.05841,
  title  = {Parallel algorithms for maximizing one-sided $\sigma$-smooth function},
  author = {Hongxiang Zhang and Yukun Cheng and Chenchen Wu and Dachuan Xu and Dingzhu Du},
  journal= {arXiv preprint arXiv:2206.05841},
  year   = {2022}
}
R2 v1 2026-06-24T11:48:13.998Z