We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:

Download:

Current browse context:

math.OC

Change to browse by:

References & Citations

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Mathematics > Optimization and Control

Title: Data-driven Optimal Computing Budget Allocation under Input Uncertainty

Abstract: In a fixed budget ranking and Selection (R&S) problem, one aims to identify the best design among a finite number of candidates by efficiently allocating the given simulation budget to evaluate design performance. Classical methods for fixed budget R&S usually assume known input distributions, which are the distributions that model the randomness in the system and drive the simulation. In this paper, we consider the practical scenario where the input distribution is unknown but can be estimated from streaming input data that arrive in batches over time. We model the R&S problem in this dynamic setting as a multi-stage problem, where the input distribution estimate is updated at each stage and a stage-wise optimization problem is formulated to allocate the simulation budget. We characterize the optimality conditions for the stage-wise budget allocation problem by applying the large deviations theory to maximize the decay rate of the probability of false selection. Based on the optimality conditions and combined with the updating of input distribution estimates, we design two sequential budget allocation procedures for R&S under streaming input data. We theoretically guarantee the consistency and asymptotic optimality of the two proposed procedures. We also demonstrate the practical efficiency through numerical experiments in comparison with the equal allocation policy and two extensions of the Optimal Computing Budget Allocation (OCBA) algorithm.
Subjects: Optimization and Control (math.OC)
Cite as: arXiv:2209.11809 [math.OC]
  (or arXiv:2209.11809v1 [math.OC] for this version)

Submission history

From: Yuhao Wang [view email]
[v1] Fri, 23 Sep 2022 18:57:40 GMT (1619kb,D)

Link back to: arXiv, form interface, contact.