Anytime Benchmarking of Forty-Five PAES-25 Configurations on Multi- and Many-Objective Variants of Leading-Ones-Trailing-Zeros Functions

What is it about?

Decisions in engineering, operations, business, healthcare, finance, energy, and many other walks of life often involve finding a compromise or tradeoff among competing interests, goals or objectives. In order to support these kind of decisions algorithmically we use formal optimization methods that can handle multiple objectives. This paper is a study of one such method (PAES-25), which was originally developed 25 years ago and has remained competitive. In order to study it in depth, we run it on some challenging benchmarks (with up to 8 objectives) and look at the time evolution of a whole dashboard of performance features while we vary the configuration of the method. This reveals some stable configurations that work well quite generally as well as some more nuanced recommendations that depend upon features of the problem being tackled.

Featured Image

Photo by Randy Laybourne on Unsplash

Photo by Randy Laybourne on Unsplash

Why is it important?

Studying the performance of algorithms over their whole time course is useful because it can reveal the best configuration(s) dependent on the available time budget available. Our study does this for a popular algorithm (cited more than 5,000 times in the literature) for the first time and using a dashboard of different measures. This reveals new observations about the best configurations of the algorithm and how to set it up.