For best experience please turn on javascript and use a modern browser!
You are using a browser that is no longer supported by Microsoft. Please upgrade your browser. The site may not present itself correctly if you continue browsing.
UvA-student Dante Niewenhuis published and defended a paper on ECTA 2022. It was a direct result of his MSc-thesis AI at the Universiteit van Amsterdam.
Dante in full swing at ECTA 2022
Dante in full swing at ECTA 2022

Benchmarks

Benchmarking is an important component of many comparison studies, in machine learning and evolutionary algorithms for example. Many of these benchmarks are continuous functions, defined as a landscape of three or more dimensions.

These functions are quite well-known, and have a reputation for being difficult to optimize. But Dante Niewenhuis questioned this premise, and used a technique known as ‘genetic programming’ to create new benchmark functions, by re-parameterizing the known ones.

Genetic programming and evolutionary algorithms

Genetic programming and evolutionary algorithms are teaching us that solutions exist that are far beyond human comprehension. Examples of these are optimal schedules for factories, industrial design, but apparently also well-known benchmark functions”, says supervisor Daan van den Berg. “But the idea that natural evolution is a good designer of organisms is religously engrained in many minds. For many people and scientists, it is impossible to imagine that machine evolution might do a much better task at designing stuff than natural evolution ever will.

Evolving the 2-dimensional ‘Goldstein-Price’ benchmark function. Bottom left is the version number of the function, bottom right its hardness
Evolving the 2-dimensional ‘Goldstein-Price’ benchmark function. Bottom left is the version number of the function, bottom right its hardness