In 2012, researchers in the genetic programming community realized that the field was being held back by reliance on toy benchmark problems [1]. By 2015, some work had been done to improve benchmarking in specialized areas such as symbolic regression [2]. However, no standardized benchmark problems had yet been developed for general program synthesis, the task of automatically generating programs that are similar in many respects to the kinds of programs that humans usually write.
This dissertation explores how optimization algorithms can become both more effective and more interpretable. It focuses on two long-standing limitations of metaheuristics: first, their frequent lack of awareness of the specific structure of the problem they are solving; and second, their opacity, which makes it difficult to understand why they behave well in some cases and poorly in others.
Q1: Which are the service, editorial, leadership, mentoring or other contributions you are most proud of? It is most rewarding for me to see former students and postdocs progressing their careers, and I regard this as most important. Over the years, I have had many students and postdocs from a wide range of countries join me in Adelaide for some time. Many stayed in Australia, and others continued their journey overseas. It is always great to catch up with some of them at GECCO and other conferences.
The SIGEVO Outstanding Contribution Award recognizes remarkable contributions to Evolutionary Computation (EC) when evaluated over a sustained period of at least 15 years. These contributions can include technical innovations, publications, leadership, teaching, mentoring, and service to the EC community. In 2025, two accomplished members of our community received this recognition: Mengjie Zhang and Frank Neumann. To celebrate these distinctions, they kindly answered our questions reflecting on their contributions and views on evolutionary computation, as well as their advice to young researchers in the field.
Why this paper is a lost gem Over the past three decades, evolutionary computation has produced a diverse array of algorithms, each adapted to specific representations or problem classes. Yet the field has long lacked a unifying theory to explain why these methods often succeed despite surface differences. In this remarkable paper, Alberto Moraglio took a decisive step toward such a unification.
The construction sector remains one of the world's major sources of greenhouse gas emissions. As of 2024, buildings and construction account for about 34% of global energy-related CO. emissions and over 32% of total energy consumption. Although a temporary drop was observed in 2020 due to the COVID-19 pandemic, emissions have since rebounded and are now stagnating or increasing slightly, driven by the expansion of the global building stock and rising demand for heating and cooling. Despite notable progress in efficiency and renewable integration, the sector's overall carbon footprint continues to reach record levels, highlighting the urgent need for more advanced optimization and design strategies.
The 2025 Genetic and Evolutionary Computation Conference (GECCO) took place in the vibrant city of Málaga, Spain from July 15th to 19th in hybrid mode. Bathed by the Mediterranean Sea, Málaga offered attendees temperate sunny weather, wonderful food, and a rich cultural life, with its historic monuments and museums. The total number of attendees was 787: 597 in-person (76%), and 190 online (24%).
Benchmarking plays an important role in understanding the strengths and weaknesses of optimization heuristics. However, data collected from one benchmarking study is not limited in its use to answering that single original research question. By sharing and re-using data, we can gain new insights into, for example, the complementarity of different algorithms. In this thesis, we start from the creation of a software framework which supports the benchmarking process, and utilize the resulting data to investigate meta-algorithmic questions, starting from algorithm configuration and moving to dynamic algorithm selection: switching between optimizers during the search process.
Visual illusions are a popular subject for entertainment, and they are also an active yet still mysterious research subject. There is no consensus on the mechanisms for many of the most popular illusions, and new types of striking visual illusions regularly "go viral" in social media. This contribution summarizes our research article, where we try to answer the question of why some static images appear to move (motion illusions) through the use of a prediction-based perception model coupled to a simple pattern generation algorithm. Our results support the hypothesis that illusory motion might be the result of perceiving the brain's own visual predictions, rather than perceiving raw visual input from the eyes. The philosophical motivation of this paper is to call attention to the untapped potential of "motivated failures", ways for artificial intelligence systems to fail as biological systems fail, as a worthy outlet for understanding the brain.
The GECCO 2025 Conference was held July 14-18 this year, once again in hybrid mode. Nearly 600 people came to Málaga, Spain to attend in person, and an additional 200 or so registered to attend virtually. The finals of the Humies competition (see www.human-competitive.org for the Call for Entries) was a plenary session on Thursday, July 17, that was attended by about 150 people. The eight finalists earlier selected by the judges (from among 15 entries this year, and based on the papers and entry forms they submitted) presented their work in 10-minute talks, or, in one case, a pre-recorded video.
Artificial Intelligence (AI) and Evolutionary Computation (EC) have long served as complementary approaches: AI learns from data, while EC explores complex solution spaces without gradients or supervision. EC has traditionally helped optimize neural networks by tuning hyperparameters, designing architectures, and evolving symbolic models. Today, this relationship is shifting. Rather than acting in support roles, AI and EC are beginning to co-design systems.
The rapid advancements in Machine Learning (ML) and Black-Box Optimisation (BBO) have led to an increased reliance on benchmarking data for evaluating and comparing algorithms across diverse domain tasks. However, the effective exploitation of this data is hindered by challenges such as syntactic variability, semantic ambiguity, and lack of standardization. In this dissertation, we address these challenges by advocating for formal semantic representation of benchmarking data through the use of ontologies. By providing standardized vocabularies and ontologies, we improve knowledge sharing and promote data interoperability across studies in ML and BBO. In the ML domain, focusing on multi-label classification (MLC), we design an ontology-based framework for semantic annotation of benchmarking data, facilitating the creation of MLCBench - a semantic catalog that enhances data accessibility and reusability. In the BBO domain, we introduce the OPTION (OPTImization algorithm benchmarking ONtology) ontology to formally represent benchmarking data, including performance data, algorithm metadata, and problem landscapes. This ontology enables the automatic integration and interoperability of knowledge and data from diverse benchmarking studies.
Embodied Artificial Intelligence (AI) refers to the integration of AI systems within a (simulated or physical) body, such as AI-driven robots. This paradigm is inspired by biological organisms, aiming to replicate the dynamic interaction between intelligent behavior and a body. However, the resemblance to biological beings is frequently a coarse approximation, raising questions about the extent to which these systems genuinely capture biological principles.
Constrained multiobjective optimization problems (CMOPs) are integral to many real-world applications. Despite significant recent advances in multiobjective optimization, there remains a lack of adequate characterization and understanding of CMOPs, which limits the development and benchmarking of optimizers. This thesis addresses this gap by investigating CMOPs from two complementary perspectives: the feature space and the performance space.
Last year's Evolutionary Computation Theory and Applications (ECTA 2024) conference took place in the beautiful city of Porto, Portugal, from November 20th to 22nd, 2024, as part of the International Joint Conference on Computational Intelligence (IJCCI). The event, organized by INSTICC and supported by ACM SIGAI, AAAI, and other esteemed institutions, brought together researchers worldwide to discuss their latest advancements.
Training agents to perform complex tasks like driving a car, mastering a video game, or controlling a robot to walk presents a significant challenge when expert demonstrations are not available. In nature, complex behaviors and characteristics can emerge through evolution, as animals adapt to their environments and problems over generations.
Neuroevolution refers to using evolutionary computation to design and optimize neural networks. This thesis entitled Neuroevolution trajectory networks: illuminating the evolution of artificial neural networks, focuses on explainability in neuroevolution. Several neuroevolution algorithms and frameworks have been proposed, which underscores the rising need for methods to evaluate and understand them and their optimisation dynamics.
In 1999 the first Genetic and Evolutionary Computation Conference (GECCO) [1] was founded as a "recombination" of the annual Genetic Programming (GP) conference [2] and the biannual International Conference on Genetic Algorithms (ICGA), which had been started in 1985 and was held every two years in the USA.
It's now over two years since Julian F. Miller sadly passed away in February 2022 and lots of development has taken place within the graph-based genetic programming (GGP) community since then.
The GECCO 2024 Conference was held July 14--18 this year, once again in hybrid mode. Nearly 400 people came to Melbourne, Australia to attend in person, and an additional 275 or so registered to attend virtually. The finals of the Humies competition (see www.human-competitive.org for the Call for Entries) was a plenary session on Wednesday, July 17, that was attended by about 150 people. The eight finalists earlier selected by the judges (from among 16 entries, and based on the papers and entry forms they submitted) presented their work in 10-minute talks, or, in one case, a pre-recorded video.
