English

Fast minimization of structured convex quartics

Optimization and Control 2018-12-27 v1

Abstract

We propose faster methods for unconstrained optimization of \emph{structured convex quartics}, which are convex functions of the form \begin{equation*} f(x) = c^\top x + x^\top \mathbf{G} x + \mathbf{T}[x,x,x] + \frac{1}{24} \mathopen\| \mathbf{A} x \mathclose\|_4^4 \end{equation*} for cRdc \in \mathbb{R}^d, GRd×d\mathbf{G} \in \mathbb{R}^{d \times d}, TRd×d×d\mathbf{T} \in \mathbb{R}^{d \times d \times d}, and ARn×d\mathbf{A} \in \mathbb{R}^{n \times d} such that AA0\mathbf{A}^\top \mathbf{A} \succ 0. In particular, we show how to achieve an ϵ\epsilon-optimal minimizer for such functions with only O(n1/5logO(1)(Z/ϵ))O(n^{1/5}\log^{O(1)}(\mathcal{Z}/\epsilon)) calls to a gradient oracle and linear system solver, where Z\mathcal{Z} is a problem-dependent parameter. Our work extends recent ideas on efficient tensor methods and higher-order acceleration techniques to develop a descent method for optimizing the relevant quartic functions. As a natural consequence of our method, we achieve an overall cost of O(n1/5logO(1)(Z/ϵ))O(n^{1/5}\log^{O(1)}(\mathcal{Z} / \epsilon)) calls to a gradient oracle and (sparse) linear system solver for the problem of 4\ell_4-regression when AA0\mathbf{A}^\top \mathbf{A} \succ 0, providing additional insight into what may be achieved for general p\ell_p-regression. Our results show the benefit of combining efficient higher-order methods with recent acceleration techniques for improving convergence rates in fundamental convex optimization problems.

Keywords

Cite

@article{arxiv.1812.10349,
  title  = {Fast minimization of structured convex quartics},
  author = {Brian Bullins},
  journal= {arXiv preprint arXiv:1812.10349},
  year   = {2018}
}
R2 v1 2026-06-23T06:56:23.257Z