English

Efficient displacement convex optimization with particle gradient descent

Machine Learning 2023-02-10 v1 Machine Learning

Abstract

Particle gradient descent, which uses particles to represent a probability measure and performs gradient descent on particles in parallel, is widely used to optimize functions of probability measures. This paper considers particle gradient descent with a finite number of particles and establishes its theoretical guarantees to optimize functions that are \emph{displacement convex} in measures. Concretely, for Lipschitz displacement convex functions defined on probability over Rd\mathbb{R}^d, we prove that O(1/ϵ2)O(1/\epsilon^2) particles and O(d/ϵ4)O(d/\epsilon^4) computations are sufficient to find the ϵ\epsilon-optimal solutions. We further provide improved complexity bounds for optimizing smooth displacement convex functions. We demonstrate the application of our results for function approximation with specific neural architectures with two-dimensional inputs.

Keywords

Cite

@article{arxiv.2302.04753,
  title  = {Efficient displacement convex optimization with particle gradient descent},
  author = {Hadi Daneshmand and Jason D. Lee and Chi Jin},
  journal= {arXiv preprint arXiv:2302.04753},
  year   = {2023}
}
R2 v1 2026-06-28T08:36:04.032Z