Newsletter of the ACM Special Interest Group on Genetic and Evolutionary Computation

Editorial

Welcome to the 4th (Winter) 2025 issue of the SIGEvolution newsletter! Our opening article and accompanying cover image give facts and figures of this year Genetic and Evolutionary Computation Conference, GECCO 2025 celebrated in Malaga, Spain. We continue with a Lost Gem contribution on Abstract Convex Evolutionary Search. We conclude with the usual announcements, forthcoming events and call for submissions. Remember to get in touch if you’d like to contribute or have suggestions for future issues of the newsletter.

Gabriela Ochoa (Editor)

Conference Overview

GECCO 2025: A Visual Analysis

Gabriela Ochoa, University of Stirling, UK, and Bogdan Filipič, Jožef Stefan Institute, Slovenia 

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%).

GECCO 2025 received 501 full paper submissions and accepted 181, resulting in a 36.1% acceptance rate. Additionally, 40 Hot-off-the-Press (HOP) presentations, 13 Late Breaking Abstracts (LBA), and 218 posters were presented during the conference. GECCO 2025 also included 35 tutorials, selected from among 58 proposals, as well as 19 workshops reflecting the most relevant topics in our field. Other events held during the conference were the Humies presentations, the Students’ workshop, Evolutionary Computation in Practice, the Job Market, and Women+@GECCO. We were informed and inspired by two Spanish keynote speakers: Maria Amparo Alonso Betanzos, University of A Coruña (Rethinking Efficiency in Machine Learning, Figure 1), and Javier Del Ser, University of the Basque Country (Evolutionary Computation as a Path to Safe, Trustworthy, and Responsible General-Purpose AI). We were also delighted by an unforgettable SIGEVO keynote from Marc Schoenauer (Evolutionary Computation: Back to the Future).

Full Paper Submissions and Acceptance Rates 

The main scientific program was organized into 14 tracks: BBSR – Benchmarking, Benchmarks, Software and Reproducibility, CS – Complex Systems, ECOM – Evolutionary Combinatorial Optimization and Metaheuristics, EML – Evolutionary Machine Learning, EMO – Evolutionary Multiobjective Optimization, ENUM – Evolutionary Numerical Optimization, GA – Genetic Algorithms, GECH – General Evolutionary Computation and Hybrids, GP – Genetic Programming, L4EC – Learning for Evolutionary Computation, NE – Neuroevolution, RWA – Real World Applications, SI – Swarm Intelligence and THEORY – Theory. 

Figure 2 shows the number of submissions (yellow bars) and acceptances (blue bars) by track, sorted by decreasing number of submissions. The tracks with the highest number of submissions (over 50) were RWA, EMO, and ECOM. The acceptance rates per track are also shown (green line), with the conference overall rate represented by a solid red line. This year, the SIGEVO Executive Board agreed on a maximum target of 40% acceptance rate for all tracks. This was achieved by most tracks, with only a few small deviations. Three of the tracks (RWA, EMO, and GA) have an acceptance rate closer to 30% than to 40%.

Figure 3 illustrates the history of submissions, acceptances, and rates since 2005. We can see that the dip in submissions observed during the COVID-19 years (2020 and 2021) is now recovering, with the last 3 years showing a stable number of submissions (close to 500) and acceptance rate (close to 36.5%). An interesting observation in this plot is the alternating ups and downs in the number of submissions following alternating locations in Europe and the United States (US), respectively. This is especially noticeable from 2005 to 2013. Hosting the conference outside the US helps to alleviate this effect, as is noticeable when the conference was hosted in Vancouver (2014), Kyoto (2018) and Melbourne (2024).

Authors of Accepted Papers by Country and Region

We now analyse the country of affiliation of all authors of accepted full papers. We adopt a colouring scheme by continental region (see legend in Figure 4), and are happy to report that all world regions are represented. Figure 4 shows the number of authors per country in decreasing order of number of authors. We can see that China is the country with the largest number of authors, followed by the United Kingdom and Germany, with the United States in 4th place. Japan is very well represented in 6th place. Portugal outnumbered larger countries in Europe like France and Spain, which is curious, as the conference took place in Spain. Brazil is the country with the largest number of authors in Latin America, outnumbering Mexico. 

Figure 5 shows another perspective of the number of authors by country of affiliation, now grouping countries by continental region, and having circle sizes proportional to the number of authors. We can see that Europe dominates, showing the largest number of countries and overall size; the largest in Europe are the UK and Germany, showing similar sizes. Asia is second, with a clear China domination, followed by Japan. America is third, where, interestingly, Brazil outnumbers both Canada and Mexico. Oceania follows with similar sizes for Australia and New Zealand. Africa is represented by 3 countries, with Tunisia showing the largest size. 

Figure 6 shows the distribution of the number of authors of accepted papers by track and continental region, using a Sankey diagram. A Sankey diagram is a type of flow diagram that allows us to visualise the flow and quantity of a resource through a system or process. In this case, the resource is the number of authors. The width of the links is proportional to the quantity they represent. The nodes, visualised as black blocks with labels, indicate the regions (on the left) and tracks (on the right). The links/flows are bands connecting the nodes. Their widths directly indicate the magnitude of the flow. A thicker link means a larger number of authors.

The figure shows that authors from European affiliations have papers in all tracks, with the largest proportion in ECOM and RWA. Authors from institutions in the Americas also have papers in all tracks, but with a different proportion, with EML, GP, and EMO being the highest. Asian authors have papers in all tracks except CS, which has authors only from Europe and the Americas. EMO is the track with the highest proportion of Asian authors. 

Co-Authorship Patterns by Region

We conclude with an analysis of co-authorship by constructing a bipartite co-authorship network. This network explicitly models the relationship between two sets of entities: authors and papers. In the context of co-authorships, a bipartite network G = (A, P, E) consists of

• A (Authors): Set of nodes representing individual researchers, with their region of affiliation as a property.
• P (Papers): Set of nodes representing papers, with their main GECCO track as a property.
• E (Edges): Edges only exist between an author node and a paper node. An edge connects an author a 𝞊 A to a paper p 𝞊 P if author a is an author of paper p. No edges exist within the set of authors or within the set of papers.

Figure 7 visualises the bipartite co-authorship network. Author nodes are filled circles, and paper nodes are outlined in black squares. Author nodes are colored by region of affiliation. The network has 112 components. The graph layout organises the components by size, with components of size 2, 3 and 4 in the first three rows. At the top left, we can see two components of size two; they are single-author papers which both happen to have European (German) affiliations. Thereafter, we can see a large number of components of sizes 3 and 4 filling the rest of the first row, as well as the second and third rows. These are papers with 2 or 3 authors; we can see that the pattern of co-authorship varies, with most components having a homogeneous colour (region), but some have mixed colours showing inter-region co-authorships.

Figure 8 provides an overview of the component sizes. We can see that a component size of 2 is rare, but the most common component sizes are linearly 3, 4, 5, 6, 7 and 8. However, the plot has a “tail,” highlighting the presence of large components, with up to 44 nodes. Remember that components have both author and paper nodes, so large sizes involve several papers with a group of authors linked by co-authorships. Looking again at Figure 7, we can appreciate the two largest components of sizes 34 and 44 in the bottom row. The size 34 component on the left side has 2 colours (red and blue), indicating a pattern of co-authorship between authors in Europe and the Americas. A closer look at this component indicates it has 23 authors, with affiliations in 10 countries within two regions; they co-authored 11 papers across 7 GECCO tracks. The largest component, with 44 nodes, contains authors from 4 regions! (all regions except Africa). This component hosts 31 authors with affiliations in 11 countries across 4 regions; together they produced 13 papers across 9 GECCO tracks. 

The animated GIF in Figure 9 reveals the identity of authors and paper tracks of the largest components (of size 10 or larger) of the bibartite co-authorship network.

Acknowledgements

Organizing a conference like GECCO is a tremendous task that relies on many people. We would like to thank all the chairs of our events: tracks, posters, workshops, student workshop, tutorials, competitions, LBA, and HOP. We also thank the organizers of Humies, Evolutionary Computation in Practice, Job Market, SIGEVO Summer School, and Women+@GECCO, as well as the members of our program committee. Many other members of the organization team deserve recognition (Figure 10 shows part of the team): the GECCO local chairs, local organization team, and dozens of student volunteers that help run the hybrid conference, as well as the chairs of all the different elements that hybrid GECCO is made of: Hybridization, Hybrid Scheduling, Proceedings, Student Affairs, Electronic Media, Publicity, Sponsorships, Sustainability, and SIGEVO Electronic Media Affairs. It is also worth mentioning Roxane Rose and Stephanie Matal of Executivevents for their hard work with the registrations and logistics, Leah Glick, Mark Montague, and Taylor Carr of the Linklings team for their support with the submission and review management system, Robert Mercado and Daniel Vogt from Whova for their assistance with the event management system, as well as Maribel Tineo, Diana Brantuas and John Otero of ACM for their organizational support. Finally, we also thank Anne Auger, Manuel López-Ibáñez, Markus Wagner, and Peter Bosman from SIGEVO for their valuable advice and guidance. 

---

Lost Gem

Abstract Convex Evolutionary Search

Yong-Hyuk Kim, Kwangwoon University in Seoul, South Korea

Alberto Moraglio (2011) Abstract Convex Evolutionary Search. In Proceedings of the 11th Workshop on Foundations of Genetic Algorithms (FOGA ’11), ACM, New York, NY, USA, 151–162.
https://doi.org/10.1145/1967654.1967668

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 key insight is striking not only in its simplicity and generality but also in its mathematical elegance. All evolutionary algorithms using geometric crossover without mutation perform a single, canonical form of search: convex search. Moraglio rigorously demonstrates that, regardless of the representation, applying selection and geometric recombination iteratively shrinks the convex hull of the population. This formulation is both formally rigorous and intuitively compelling, revealing deep structural commonalities that had previously remained hidden beneath the diversity of representations. The result is a nested sequence of convex hulls that fully describes the search dynamics in a representation-independent way.

Beyond its formal elegance, the paper explores what classes of fitness landscapes are inherently compatible with convex search. Moraglio proposes that concave and quasi-concave landscapes, generalized beyond Euclidean spaces to combinatorial and discrete domains, allow evolutionary algorithms to make steady progress even without explicit selection pressure. This framework conceptually connects geometric crossover with the empirically observed “big valley” structure in many optimization problems. Notably, it also clarifies the limits of metaphor in evolutionary computation, showing that while the language of natural selection often blurs the line between heuristic analogy and literal modeling, the search dynamics can be described with mathematical precision independent of biological narratives. This distinction contributes to a deeper understanding of what it means for evolutionary algorithms to be models rather than mere metaphors.

Why it was overlooked

Despite its originality, this paper has not been widely cited or broadly adopted. Since its publication, it has been cited approximately 30 times excluding self-citations. The citation record shows that a small number of researchers have continued to reference it intermittently over more than a decade, but its influence has not grown steadily over time. Many of these citations appear in conceptual discussions or theoretical surveys rather than in applied studies that build directly on the framework.

One reason is that the level of abstraction is quite high, making it difficult for practitioners to engage with the material. The paper primarily presents a theoretical construct without offering clear design guidelines or empirical validation that would help demonstrate its practical relevance. In addition, some in the community perceived it mainly as a contribution to clarifying the conceptual boundaries of evolutionary computation. It helped distinguish rigorously defined search processes from evocative but potentially misleading metaphors of biological evolution, rather than serving as an immediately applicable design framework. As a result, the proposal of a unifying perspective did not develop into a broader body of follow-up research or become integrated into mainstream evolutionary computation practice.

Why it deserves rediscovery

Today, the field is rapidly converging on exactly the kinds of questions Moraglio’s work anticipated. The rise of automated algorithm configuration, AutoML, and meta-evolutionary methods has created an urgent need for universal models that can formally describe and predict operator behavior across diverse representations. In many cases, conventional empirical benchmarks or isolated performance comparisons no longer suffice to explain why certain search strategies succeed or fail.

In particular, neuroevolution and combinatorial optimization increasingly reveal that the geometric properties of representations and operators critically shape optimization performance. Yet there are still few theoretical frameworks that can capture these relationships in a rigorous, representation-independent way. Moraglio’s framework is unique in this respect, providing a principled lens to interpret all geometric crossover operators as instances of a single, unifying concept of convex search.

Moreover, the paper offers one of the rare formal explanations for why evolutionary algorithms often perform robustly on fitness landscapes with a “big valley” structure or a global convex trend. In doing so, it also helps clarify the philosophical status of emergent complexity in evolutionary algorithms. By showing that apparent novelty and unpredictability arise within fully specified mathematical constraints, it informs contemporary debates about whether such algorithms genuinely exhibit emergence or merely produce epistemically bounded surprises. This perspective could be highly valuable for emerging areas such as AutoML and hyperparameter optimization, where predicting problem-algorithm fit, designing search strategies in advance, and communicating interpretability are becoming essential.

Revisiting this work is not merely a retrospective exercise. It is an opportunity to establish a theoretical foundation that addresses some of the most pressing challenges in contemporary evolutionary computation. At a time when the field faces growing demands for conceptual clarity and algorithmic transparency, there has never been a more timely moment to reassess and build upon this contribution.

Disclosure: I coauthored research with Dr. Moraglio more than ten years ago. I have no ongoing collaborations or professional ties, and this recommendation is made solely based on the merits of the work.

About the Author

---

Announcements

Summary of the ZEBAI Project

Antonio LaTorre, Universidad Politécnica de Madrid (UPM)

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 CO₂ 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 ZEBAI project proposes the development of an advanced methodology for the automated design of Zero Emission Buildings (ZEBs) using Evolutionary Algorithms (EAs) and other Artificial Intelligence techniques. Its main goal is to optimize building performance from the early design stages, considering multiple and often conflicting objectives such as energy efficiency, cost, thermal comfort, daylight availability, and environmental impact. By integrating simulation tools like EnergyPlus or Radiance with evolutionary optimization methods, ZEBAI aims to explore vast design spaces and identify high-performance architectural solutions that would be difficult to achieve through traditional design approaches.

In this context, Evolutionary Algorithms are used as the core optimization engine to evolve populations of design alternatives iteratively. Each candidate solution—representing a combination of building materials for each of the components of its envelope (windows, walls, roofs), HVAC systems, renewable energy equipments—is evaluated through detailed performance simulations. Evolutionary Algorithms progressively improve the population toward optimal trade-offs, effectively balancing energy demand reduction, occupant comfort, cost, etc. This approach enables designers to discover innovative configurations that optimize the aforementioned objectives according to their preferences taking into account more options than an architect normally considers when designing by hand.

Moreover, the ZEBAI project emphasizes the creation of a flexible and scalable computational framework capable of integrating heterogeneous simulation tools and data sources. Through modular interfaces and workflow automation, the platform facilitates interoperability between modeling environments and allows multi-objective optimization of buildings, regardless of their location or nature. The ultimate vision of ZEBAI is to support architects, engineers, and planners in the transition toward carbon-neutral built environments, leveraging artificial intelligence to make sustainable design both efficient and accessible.

To complement the computational framework, the ZEBAI project also includes the development of a user-friendly graphical interface that allows designers to interact intuitively with the optimization process. This interface provides visual feedback on design performance, evolutionary progress, and trade-offs between objectives. By simplifying access to complex optimization workflows and simulation data, the interface enables architects and engineers—regardless of their programming expertise—to define parameters, constraints, or regulations, monitor results, and explore optimal design alternatives. Ultimately, this component ensures that ZEBAI’s advanced AI-based optimization capabilities are both accessible and practical for use in real-world building design practice.

About the Author

---

Journal Latest Issues

---

---

Call For Submissions

The Genetic and Evolutionary Computation Conference (GECCO) presents the latest high-quality results in genetic and evolutionary computation since 1999. Topics include: genetic algorithms, genetic programming, swarm intelligence, complex systems, evolutionary combinatorial optimization and metaheuristics, evolutionary machine learning, learning for evolutionary computation, evolutionary multiobjective optimization, evolutionary numerical optimization, neuroevolution, real-world applications, theory, benchmarking, reproducibility, hybrids and more.

Detailed Call for Papers.

Important Dates

---

The International Conference on Parallel Problem Solving From Nature (PPSN) is a biannual open forum fostering the study of natural models, iterative optimization heuristics, machine learning, and other artificial intelligence approaches. PPSN was originally designed to bring together researchers and practitioners in the field of Natural Computing, the study of computing approaches that are gleaned from natural models. Today, the conference series has evolved and welcomes works on all types of iterative optimization heuristics. Notably, we also welcome submissions on connections between search heuristics and machine learning or other artificial intelligence approaches. Submissions covering the entire spectrum of work, ranging from rigorously derived mathematical results to carefully crafted empirical studies, are invited.

Detailed Call for Papers.

---

For Human-Competitive Results – Produced by Genetic and Evolutionary Computation

To be held as part of the Genetic and Evolutionary Computation Conference (GECCO)
July 13-17, 2026 (Monday – Friday) San Jose, Costa Rica (Hybrid)

Detailed Call for Entries

Entries are hereby solicited for awards totaling $10,000 for human-competitive results that have been produced by any form of genetic and evolutionary computation (including, but not limited to genetic algorithms, genetic programming, evolution strategies, evolutionary programming, learning classifier systems, grammatical evolution, gene expression programming, differential evolution, genetic improvement, etc.) and that have been published in the open, reviewed literature between the deadline for the previous competition and the deadline for the current competition.

Important Dates

• Friday, May 29, 2026: Deadline for entries (consisting of one TEXT file, PDF files for one or more papers, and possible “in press” documentation). Please send entries to goodman at msu dot edu
• Friday, June 12, 2026: Finalists will be notified by e-mail
• Friday, June 26, 2026: Finalists must submit a 10-minute video or, if presenting in person, their slides, to goodman at msu dot edu.
• July 13-17, 2026 (Monday – Friday): GECCO conference (the schedule for the Humies session is not yet final, so please check the GECCO program as it is updated for the time of the Humies session). GECCO will be in hybrid mode, so the finalists may present their entry in person or on video.
• Friday, July 17, 2026: Announcement of awards at the plenary session of the GECCO conference.

About this Newsletter