She develops algorithms that outperform leading optimization techniques and can be applied in many fields. For this work, Niki van Stein and her team have won the GECCO Humies Award. 'With our method, we can create and refine algorithms for a wide range of applications-algorithms that are better than the ones we currently have.'
The GECCO Humies Awards recognize researchers who demonstrate that AI-developed methods can achieve better results than human experts working by hand. The Leiden team-Niki van Stein, Thomas Bäck, Haoran Yin, and Anna Kononova-received the silver award for their LLaMEA: the Large Language Model Evolutionary Algorithm. 'We study the combination of so-called evolutionary methods and large language models, like ChatGPT. Our goal is to evolve algorithms and automatically improve them,' Van Stein explains.
A method with wide-ranging applications
The researchers found that they can design algorithms that perform better than the best hand-crafted approaches. 'With this method, we can solve all kinds of complex problems,' says Van Stein. 'For example, in photonics, where you can fine-tune materials so they have exactly the properties you want. But our method goes far beyond that: you can also use it to make AI systems smarter, improve mathematical models, and develop new computational techniques.'
Research with impact-and a boost in morale
Winning the award means Van Stein can take her research even further. 'For instance, we can now work with multiple large language models, which are very expensive to use. Our work is making an impact and getting noticed by other researchers. That kind of recognition is a real boost for us.'
