Salvato in:
Dettagli Bibliografici
Autori principali: Bollapragada, Raghu, Karamanli, Cem, Wild, Stefan M.
Natura: Preprint
Pubblicazione: 2024
Soggetti:
Accesso online:https://arxiv.org/abs/2404.11893
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866916211941441536
author Bollapragada, Raghu
Karamanli, Cem
Wild, Stefan M.
author_facet Bollapragada, Raghu
Karamanli, Cem
Wild, Stefan M.
contents In this paper, we present a novel derivative-free optimization framework for solving unconstrained stochastic optimization problems. Many problems in fields ranging from simulation optimization to reinforcement learning involve settings where only stochastic function values are obtained via an oracle with no available gradient information, necessitating the usage of derivative-free optimization methodologies. Our approach includes estimating gradients using stochastic function evaluations and integrating adaptive sampling techniques to control the accuracy in these stochastic approximations. We consider various gradient estimation techniques including standard finite difference, Gaussian smoothing, sphere smoothing, randomized coordinate finite difference, and randomized subspace finite difference methods. We provide theoretical convergence guarantees for our framework and analyze the worst-case iteration and sample complexities associated with each gradient estimation method. Finally, we demonstrate the empirical performance of the methods on logistic regression and nonlinear least squares problems.
format Preprint
id arxiv_https___arxiv_org_abs_2404_11893
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Derivative-Free Optimization via Adaptive Sampling Strategies
Bollapragada, Raghu
Karamanli, Cem
Wild, Stefan M.
Optimization and Control
In this paper, we present a novel derivative-free optimization framework for solving unconstrained stochastic optimization problems. Many problems in fields ranging from simulation optimization to reinforcement learning involve settings where only stochastic function values are obtained via an oracle with no available gradient information, necessitating the usage of derivative-free optimization methodologies. Our approach includes estimating gradients using stochastic function evaluations and integrating adaptive sampling techniques to control the accuracy in these stochastic approximations. We consider various gradient estimation techniques including standard finite difference, Gaussian smoothing, sphere smoothing, randomized coordinate finite difference, and randomized subspace finite difference methods. We provide theoretical convergence guarantees for our framework and analyze the worst-case iteration and sample complexities associated with each gradient estimation method. Finally, we demonstrate the empirical performance of the methods on logistic regression and nonlinear least squares problems.
title Derivative-Free Optimization via Adaptive Sampling Strategies
topic Optimization and Control
url https://arxiv.org/abs/2404.11893