This year’s Genetic and Evolutionary Computation Conference (GECCO 2025) in Málaga was a complete success. Our research group was represented with three contributions, all of which attracted significant interest.
Best Paper Award for TSGP
We are very proud of the Best Paper Award for the contribution “Transformer Semantic Genetic Programming for Symbolic Regression” by Philipp Anthes, Dominik Sobania, and Franz Rothlauf.
The paper presents an innovative approach that solves symbolic regression problems by integrating transformer-based neural networks into genetic programming. By taking solution behavior into account, the method can produce compact models with strong predictive performance after only a few iterations.
Evolutionary Image Generation with ImageBreeder
Star Wars goes GECCO – May the fo
rce be with your image generation!
Martin Briesch and Dominik Sobania presented ImageBreeder, a novel approach to evolutionary image generation.
The system combines diffusion models with evolutionary algorithms to algorithmically optimize the visual quality of generated images. The project attracted significant attention and shows promising potential for fully autonomous image optimization—without any human feedback.
Benchmarking Selection Methods in Genetic Programming
Alina Geiger presented her recently published benchmark study in Transactions on Evolutionary Learning and Optimization.
The study compares the performance of modern Lexicase-based selection methods (ε‑Lexicase, Batch‑Lexicase, Plexicase) with traditional methods like Tournament Selection and Fitness-Proportionate Selection—each combined with various downsampling strategies. The results offer valuable insights for choosing and combining selection and downsampling strategies in genetic programming.
We sincerely congratulate everyone involved on their contributions and are already looking forward to GECCO 2026!
We’re happy to share that our team won the Best Poster Award at EvoStar 2025 in Trieste, Italy!
Our paper, “Was Tournament Selection All We Ever Needed? A Critical Reflection on Lexicase Selection”, took a fresh look at how we select individuals in genetic programming. We found that tournament selection, when combined with down-sampling, performs similar to lexicase selection while being more efficiently!
Thanks to Alina Geiger and Martin Briesch who represented us in Trieste and presented the poster.
You can check out the paper here if you’re curious about the details.
Congratulations to Clarissa Krämer (MSc.), Susanne Schmitt (MSc.), and Prof. Dr. Franz Rothlauf on receiving the Best Student Paper Award at the Medical Informatics Europe (MIE) Conference in Glasgow, Scotland, from May 19-21, 2025.
Their award-winning paper, "Using Machine Learning for the Fusion of Tumor Records on a Real-World Dataset" focuses on consolidating multiple records describing the same tumor into a single record for each tumor. They used an artificial neural network and compared its performance with that of a deterministic, rule-based approach. They used a tabular, real-world dataset that included colorectal, breast, and prostate cancer.
Key findings were that
- Artificial Neural Networks outperform the deterministic rule-based approach.
- The performance depends on the number of features and the distribution of data.
- The predictive performance increases with a lower number of categories within a variable and a more balanced dataset.
Read the paper here: https://pubmed.ncbi.nlm.nih.gov/40380540/
Recent Publications
Books
