Cartesian Genetic Programming - From foundations to recent developments and applications
Description
Cartesian Genetic Programming (CGP) is a form of genetic programming originally developed by Julian Miller and Peter Thomson in 1997. Its classic form utilizes an integer-based genetic representation of a program decoded as a directed graph. Numerous studies have demonstrated its efficiency compared to other genetic programming techniques. Since its inception, CGP has been refined with features such as automatically defined functions, self-modification operators and multi-modal capacities. Quite recently, progress has been made in the extension of CGP through the use of techniques from genetic algorithms, crossover operators and multi-objective optimization. CGP has also showed its capacity to step up to real-world problems with recent applications in optimal control, ATARI game playing and biomedical data analysis. This tutorial was given for the first time at GECCO 2025 and was very well-received. As last year, we propose to cover the foundational techniques, advanced developments, and applications across various problem domains. Specifically, we propose to present:
- The foundational elements of Cartesian Genetic Programming (encoding, decoding, evolution);
- Advances based on the classic CGP model (e.g., self-modifying operators, mixed-type, multi-modal extensions);
- Evolutionary techniques used for evolving CGP-based genomes, from the original 1+$\lambda$ evolution strategy to recent crossover and genetic algorithm-based approaches;
- Recent real-world applications of CGP and their positive impacts on interpretability and data efficiency within these application domains.
Organizers
RWTH Aachen University, Germany
Roman Kalkreuth is currently an assistant professor at RWTH Aachen University in Germany. Primarily, his research focuses on the analysis and development of algorithms for graph-based genetic programming. From 2015 until 2022, he was a research associate of the Computational Intelligence Research Group of Professor Günter Rudolph at TU Dortmund University (Germany). Roman Kalkreuth defended his PhD thesis in July 2021 and then took up a postdoctoral researcher position within Professor Rudolph’s group. From October 2022 to June 2023, he worked in the Natural Computing Research Group of Professor Dr. Thomas Bäck at the Leiden Institute of Advanced Computer Science, which is part of Leiden University. He joined Laboratoire d’Informatique de Paris 6 (LIP6) of Sorbonne University in Paris as a postdoctoral researcher under supervision of Carola Doerr from June 2023 until March 2024. He then took up an assistant professor position at RWTH Aachen University, which started in April 2024.
Université Toulouse Capitole, IRIT - CNRS UMR5505, Institut Universitaire de France, France
Sylvain Cussat-Blanc is a Professor at the University of Toulouse Capitole in France, where he leads a research group specializing in Artificial Life, Evolutionary Algorithms, and their applications to biomedical data. His early research focused on Evo-Devo systems controlled by gene regulatory networks, with applications in modular robotics and agent control. He also developed new approaches to model complex biological systems using multi-agent systems. Over the past decade, he has been developing a research project on Cartesian Genetic Programming and innovative methods for biomedical image processing, aiming to develop an alternative to deep neural networks that produces more interpretable and trustable analysis pipelines.
ISAE-SUPAERO, University of Toulouse, France
Dennis G. Wilson is an Assistant Professor of AI and Data Science at ISAE-SUPAERO in Toulouse, France. He obtained his PhD at the Institut de Recherche en Informatique de Toulouse (IRIT) on the evolution of design principles for artificial neural networks. Prior to that, he worked in the Anyscale Learning For All group in CSAIL, MIT, applying evolutionary strategies and developmental models to the problem of wind farm layout optimization. His current research focuses on genetic programming, neural networks, and the evolution of learning.