Black-box optimization is the problem of finding the value of the parameters that minimizes a certain objective function where no analytical formulation of this function nor its derivative are available as their evaluation is only possible through a costly run of an application. Several algorithms exist in the literature that target this problem, among which are the surrogate-based black-box optimization algorithms. Their main idea is to build a cheap to evaluate model of the objective function and try to optimize this model instead of the original function. Our work is a generalization of such methods. Instead of tuning an application on a given problem, we aim at tuning it on possibly any problem it can encounter. To this end, we rely on the multitask learning and transfer learning frameworks, as they have proven to be useful in the field of machine learning when additional knowledge is available to help a prediction task. We show comparative results with state-of-the-art autotuning techniques for the tuning of computer applications. We explain how our approaches can be more suitable than some state-of-the-art autotuners for the tuning of applications in general and of expensive exascale applications in particular.

Black-box optimization of expensive functions through ML models

Fall 2022 abstracts, posters, and presentations will launch in mid-September.

ACS Fall 2022

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.