This special issue presents some of the most innovative work presented at the most important events on Genetic Programming (GP) in 2023, the European Conference on Genetic Programming (EuroGP) and the GP track at the Genetic and Evolutionary Computation Conference (GECCO). These conferences have proven to be pillars of the GP research community and consistently provide a platform for presenting groundbreaking advances and diverse applications of GP.

EuroGP, the world’s only conference dedicated exclusively to GP, took place in Brno, Czech Republic, in 2023 and was chaired by Gisele L. Pappa and Mario Giacobini. Out of the 38 submitted articles, 14 were selected for presentation. The Genetic and Evolutionary Computation Conference (GECCO), the world’s leading event in evolutionary computation and the flagship event of ACM SIGEVO, was held in Lisbon, Portugal. The GP track of GECCO 2023, chaired by Ting Hu and Domagoj Jakobovic, maintained its reputation as one of the largest and most dynamic tracks with 16 manuscripts selected for presentation among the 49 submissions.

This special issue contains the extended versions of some of the highest-rated and most promising articles presented and discussed at these events. Careful consideration was given to selecting works that were scientifically sound, original, and had the potential to impact the GP field significantly. Authors were encouraged to expand their contributions by either deepening their investigation of the proposed methods or broadening the scope of their applications. Each article underwent a thorough peer review process, including re-evaluation by the original conference reviewers and additional experts external to the conferences’ program committees. This ensures that the final collection of articles not only meets but exceeds the standards of the conferences and this journal.

The articles in this issue reflect the diversity and vitality of contemporary GP research. They cover topics ranging from theoretical insights into GP to the development of new operators and applications in real-world computing environments. Together, they offer a snapshot of the current state of the art in GP research in 2023 and provide inspiration for future research.

The first paper in this issue originates from work published at the GECCO 2023 GP track. In this paper by Nadizar et al. the authors combine Geometric Semantic Genetic Programming (GSGP) and Linear Scaling (LS) with a Darwinian and a Lamarckian evolutionary approach. The reasons for combining GSGP and LS are explained with a detailed theoretical background to both techniques. The two evolutionary approaches—Darwinism and Lamarckism—are described in detail and the expected results are outlined. A thorough experimental evaluation of the proposed methods is performed for the symbolic regression task using hand-crafted benchmarks and real-world problems. The authors analyze the results in terms of performance and generalization ability and also make comments on the convergence speed of the algorithms. Their results indicate a remarkable improvement in standard GP performance across different scenarios, which is evident in both the training and test sets. However, the authors also note that the integration of LS makes the GSGP more prone to overfitting, especially when dealing with challenging data, an issue that should be addressed in further research.

The second paper, by Leite and Schoenauer, comes from original work published in EuroGP. The authors introduce semantic boosting regression, a new approach that uses a memetic semantic GP for symbolic regression as a weak learner for a boosting algorithm. The fusion of the memetic and semantic approaches improved the exploration and exploitation capabilities inherent to GP and identified concise symbolic expressions that maintain interpretability without compromising the expressive power of symbolic regression. Experimental results demonstrate the approach has equal or better performance than other state-of-the-art evolutionary-based techniques while helping tackle the bloating issue in GP and improving its generalization capabilities.

Crary et al. author the third paper, also originally published at EuroGP, where they present an efficient implementation of the evaluation phase of tree-based GP in field-programmable gate array (FPGA). This is interesting as FPGAs can, in certain cases, leverage increased levels of both data and function parallelism and superior power/energy efficiency compared to general-purpose CPU/GPU systems. The proposed single-FPGA accelerator was slower on average than the state-of-the-art Operon tool when executing on the same 2-processor, 28-core CPU system, but better in performance-per-watt. The authors also propose further ways to improve the proposed implementation so it generates less costly GP runs, similar to how GPUs unlocked the power of deep learning during the past years.

We would like to thank the authors for their exceptional contributions and the reviewers for their dedication and expertise in ensuring the quality of this special issue. We also thank the editors-in-chief, Lee Spector and Leonardo Trujillo, for their advice, patience and support, and the editorial team at Springer for their invaluable help. Finally, we thank the GP community for their continued enthusiasm and innovation that makes these highlights possible. We hope that this special issue will inspire researchers and practitioners alike and stimulate new ideas and collaborations that will further advance the field of genetic programming.