English

Adaptive regularization with cubics on manifolds

Optimization and Control 2020-05-19 v4

Abstract

Adaptive regularization with cubics (ARC) is an algorithm for unconstrained, non-convex optimization. Akin to the popular trust-region method, its iterations can be thought of as approximate, safe-guarded Newton steps. For cost functions with Lipschitz continuous Hessian, ARC has optimal iteration complexity, in the sense that it produces an iterate with gradient smaller than ε\varepsilon in O(1/ε1.5)O(1/\varepsilon^{1.5}) iterations. For the same price, it can also guarantee a Hessian with smallest eigenvalue larger than ε1/2-\varepsilon^{1/2}. In this paper, we study a generalization of ARC to optimization on Riemannian manifolds. In particular, we generalize the iteration complexity results to this richer framework. Our central contribution lies in the identification of appropriate manifold-specific assumptions that allow us to secure these complexity guarantees both when using the exponential map and when using a general retraction. A substantial part of the paper is devoted to studying these assumptions---relevant beyond ARC---and providing user-friendly sufficient conditions for them. Numerical experiments are encouraging.

Keywords

Cite

@article{arxiv.1806.00065,
  title  = {Adaptive regularization with cubics on manifolds},
  author = {Naman Agarwal and Nicolas Boumal and Brian Bullins and Coralia Cartis},
  journal= {arXiv preprint arXiv:1806.00065},
  year   = {2020}
}

Comments

48 pages, 3 figures

R2 v1 2026-06-23T02:15:18.266Z