English

A Simple Method for Convex Optimization in the Oracle Model

Optimization and Control 2022-03-14 v3 Data Structures and Algorithms

Abstract

We give a simple and natural method for computing approximately optimal solutions for minimizing a convex function ff over a convex set KK given by a separation oracle. Our method utilizes the Frank--Wolfe algorithm over the cone of valid inequalities of KK and subgradients of ff. Under the assumption that ff is LL-Lipschitz and that KK contains a ball of radius rr and is contained inside the origin centered ball of radius RR, using O((RL)2ε2R2r2)O(\frac{(RL)^2}{\varepsilon^2} \cdot \frac{R^2}{r^2}) iterations and calls to the oracle, our main method outputs a point xKx \in K satisfying f(x)ε+minzKf(z)f(x) \leq \varepsilon + \min_{z \in K} f(z). Our algorithm is easy to implement, and we believe it can serve as a useful alternative to existing cutting plane methods. As evidence towards this, we show that it compares favorably in terms of iteration counts to the standard LP based cutting plane method and the analytic center cutting plane method, on a testbed of combinatorial, semidefinite and machine learning instances.

Keywords

Cite

@article{arxiv.2011.08557,
  title  = {A Simple Method for Convex Optimization in the Oracle Model},
  author = {Daniel Dadush and Christopher Hojny and Sophie Huiberts and Stefan Weltge},
  journal= {arXiv preprint arXiv:2011.08557},
  year   = {2022}
}

Comments

minor changes; to appear at IPCO 2022

R2 v1 2026-06-23T20:18:41.616Z