Generating data-driven continuous optimization problems for benchmarking

The School of ITEE is hosting the following HDR milestone seminar

Generating data-driven continuous optimization problems for benchmarking

Speaker: Sara Hajari
Host: A/Prof Marcus Gallagher

Abstract: Metaheuristic optimization algorithms are typically applied in black-box problem scenarios, where no strong assumptions are made about the problems. Given a large number of existing algorithms, there has been an increasing focus on the need for expanding standard benchmarking practices and problem sets, to get an accurate under-standing of the empirical performance of these algorithms as well as matching between algorithms and problems.

In this research, a detailed exploration of problem instance generation is carried out, and possible ways this approach can be used in benchmarking practice are discussed. For instance, usage of data clustering as a class of problems that can be used as a source of optimization problems for benchmarking algorithms. In addition, a method to compare problem features to investigate their efficiency is proposed.

The methodology will be then applied to a real-world Facility-Location problem. This research aims to advance understanding of the nature of optimization problem search spaces.