Michigan State University, USA

GPU computing has been one of the keys to success for deep learning models in recent years, allowing for massive scalability and high-speed performance. Currently, genetic programming systems primarily run using CPUs, and, to date, GPU technologies have provided only moderate advantages. This has held back the growth and adoption of genetic programming systems, since the CPU implementations do not meet the scale and performance expectations required for many modern applications of machine learning. The team at Noblis has allocated significant resources in the last few years to adapt genetic programming for effectively leveraging NVIDIA GPU architectures with the aim of achieving massive performance gains. Specifically, this approach has enabled the use of much larger population sizes (on the order of millions to tens of millions of individuals) while keeping generation runtimes on the scale of seconds. This high-speed performance and ability to explore genetic programming setups - unachievable when limited by CPU performance - could unlock the opportunity to solve complex problems using genetic programming and to compete properly with the now dominant deep-learning technologies. Noblis has incorporated these advances in GPU-based genetic programming into the open-source Beagle framework. Our tutorial will discuss the principles of genetic programming on GPUs, previous attempts at efficiency gains, and the key lessons and new techniques for leveraging GPUs for genetic programming from Beagle. In addition, the team plans to offer a guided tutorial of the Beagle framework so all attendees can learn how the tool leverages GPUs and how they can use Beagle for high-speed GPU-based genetic programming to enhance their research and solve new industry and scientific problems.

Department of Computer Science and Engineering, Michigan State University, East Lansing, MI, USA

Wolfgang Banzhaf is the John R. Koza Chair for Genetic Programming, the first endowed chair dedicated to Evolutionary Computation in the United States, and a professor in the Department of Computer Science and Engineering at Michigan State University, East Lansing, USA. His research interests are in the field of bio-inspired computing, notably evolutionary computation and complex adaptive system, and in particular genetic programming and artificial life. Formerly a professor at the Technical University of Dortmund, Germany (1993-2003), Memorial University of Newfoundland in Canada (2003-2016) and now Michigan State University, he is the (co-)author of more than 300 scientific contributions and 7 patents. His books and edited volumes include “Genetic Programming – An Introduction” (1998), “Linear Genetic Programming” (2007), “Artificial Chemistries” (2015) and most recently “Advances in Linear Genetic Programming” (to appear mid 2026). He served as treasurer and later chair of ACM SIGEVO, founding editor-in-chief of the SpringerNature journal ‘Genetic Programming and Evolvable Machines’ and was the first editor-in-chief of the annual GECCO conference proceedings. He won the Intl. Society for Genetic and Evolutionary Computation (now ACM SIGEVO) Senior Fellow Award in 2003, the EvoStar Award for major contributions to Evolutionary Computation in Europe in 2007 and the Intl. Society for Artificial Life (ISAL) Lifetime Achievement Award in 2022.

Noblis, USA

Ilya Basin is a Director of Software Engineering and a Computer Science Fellow at Noblis, Inc. Noblis is a Washington, D.C. metro area nonprofit science & technology organization that solves challenges across the defense, homeland security, civil, intelligence, and law enforcement sectors. Ilya is the primary creator behind the open-source Beagle framework, which is one of several open-source frameworks he has developed. Ilya holds five U.S. patents related to energy management systems, massive multi-layer simulations, and evolutionary computing. His interests include unconventional uses of GPUs for general purpose tasks, the creation of DSL compilers and interpreters, and evolutionary computing.