Vorimise

Vorimise is a term used in computational optimization to describe a family of algorithms that couple Voronoi diagram partitioning with adaptive sampling and local objective estimation. The method searches complex spaces by dividing the domain into regions around a set of seed points, and directing evaluation and refinement where the projected gain or uncertainty is highest. The name blends Voronoi diagrams with the suffix -imise, suggesting a systematic process of optimization.

Origin and scope: The concept emerged in discussions of surrogate-based optimization and adaptive meshing in the early 2010s, and has since appeared in theoretical papers as a framework for spatially aware search.

Core approach: The algorithm starts with an initial set of seeds. The objective function is evaluated at seeds; a Voronoi diagram is constructed from the seeds, assigning each region to its seed. Within each cell, a local model or surrogate estimates the objective, and a criterion (for example expected improvement or local variance) guides the placement of new samples. Seeds are updated accordingly, and the cycle repeats.

Variants and relation: Vorimise subsumes several strategies, including coarse-to-fine refinement and adaptive mesh methods. It is related to, but distinct from, Bayesian optimization and swarm-based methods, in that it emphasizes explicit spatial partitioning of the search domain.

Applications and limitations: It has been proposed for engineering design optimization, hyperparameter tuning, robotics planning, and environmental modeling. Limitations include computational cost of constructing high-dimensional Voronoi diagrams, sensitivity to initial seeds, and performance degradation as dimensionality grows without careful implementation.

See also: Voronoi diagram; surrogate model-based optimization; Bayesian optimization; adaptive mesh refinement.