Abstract:Optimizing an expensive, black-box function <math>f(·)</math> is challenging when its input space is high-dimensional. Sequential design frameworks first model <math>f(·)</math> with a surrogate function and then optimize an acquisition function to determine input settings to evaluate next. Optimization of both <math>f(·)</math> and the acquisition function benefit from effective dimension reduction. Global variable selection detects and removes input variables that do not affect <math>f(·)</math> across the input space. Further dimension reduction may be possible if we consider local variable selection around the current optimum estimate. We develop a sequential design algorithm called sequential optimization in locally important dimensions (SOLID) that incorporates global and local variable selection to optimize a continuous, differentiable function. SOLID performs local variable selection by comparing the surrogate's predictions in a localized region around the estimated optimum with the p alternative predictions made by removing each input variable. The search space of the acquisition function is further restricted to focus only on the variables that are deemed locally active, leading to greater emphasis on refining the surrogate model in locally active dimensions. A simulation study across multiple test functions and an application to the Sarcos robot dataset show that SOLID outperforms conventional approaches. <a class="ext-link" href="https://doi.org/10.1080/00401706.2020.1714738">Supplementary materials</a> for this article are available online.

Optimal space-filling design for symmetrical global sensitivity analysis of complex black-box models

A-optimal designs for non-parametric symmetrical global sensitivity analysis

Optimal Shape Design of 3-D Nonlinear Electromagnetic Devices Using Parameterized Design Sensitivity Analysis

Robust 3-D Shape Optimization of Electromagnetic Devices by Combining Sensitivity Analysis and Adaptive Geometric Parameterization

Nested symmetrical Latin hypercube designs

Sensitivity Analysis for Computationally Expensive Models using Optimization and Objective-oriented Surrogate Approximations

Reliability-oriented global sensitivity analysis using subset simulation and space partition

An Adaptive Optimal Strategy Based on the Combination of the Dynamic-Q Optimization Method and Response Surface Methodology

A Fast Global Optimal Technique Based On Combinations Of Improved Radial Basis Functions And Tabu Search Method

A Global Optimization Algorithm Based on C-1 Piecewise Response Surface Patches

Developments of an Efficient Global Optimal Design Technique – a Combined Approach of MLS and SA Algorithm

Ensemble of surrogate based global optimization methods using hierarchical design space reduction

Novel Performance-Oriented Tolerance Design Method Based on Locally Inferred Sensitivity Analysis and Improved Polynomial Chaos Expansion

Hybrid algorithms for generating optimal designs for discriminating multiple nonlinear models under various error distributional assumptions

Sequential Optimization in Locally Important Dimensions

Sensitivity Analysis of Expensive Black-Box Systems Using Metamodeling

A bilayer optimization strategy of optimal sensor placement for parameter identification under uncertainty

Approximately optimal spatial design: How good is it?

Identification Of Universally Optimal Circular Designs For The Interference Model

Mixed-Variable Global Sensitivity Analysis For Knowledge Discovery And Efficient Combinatorial Materials Design

Efficient global sensitivity analysis method for dynamic models in high dimensions