Tracks

BBSR - Benchmarking, Benchmarks, Software, and Reproducibility

Description

The Benchmarking, Benchmarks, Software, and Reproducibility track welcomes submissions that touch on all aspects of reproducibility, benchmarking, and software of genetic and evolutionary computation methods. In particular, we welcome submissions on the following topics:

• Benchmarking methodologies for assessing the performance of evolutionary algorithms and related optimization techniques,
• Benchmark problems and toolboxes for evaluating evolutionary computation methods or enabling the training of meta-learning techniques for these,
• Statistical analysis and visualization techniques for understanding problem spaces or the performance and behavior of optimization techniques, including instance space analysis and landscape analysis,
• Reproducibility studies that rigorously replicate published experiments with a substantial shift in confidence in the results of the original study,
• Innovative software for deploying, evaluating, developing, or teaching genetic and evolutionary computation in original and unique ways.

This is a non-exhaustive list, and we invite the authors to get in touch with the track chairs if in doubt about the suitability of their submission to this track.

Requirements for reproducibility studies

For reproducibility studies, the reasons for the new findings must be clearly explained in order to ensure a meaningful and distinct contribution from the original study (e.g. different benchmarks, application scenarios, technical or implementation differences). The submission must follow the highest reproducibility standards by providing all implementation details, input data, parameters and hardware specifications. All artifacts must be made available in a public repository upon submission and must remain available after publication. The submission must also follow the usual standards in terms of plagiarism.
Of particular interest are replicability studies, defined as follows by the
ACM: "The measurement can be obtained with stated precision by a different team, a different measuring system, in a different location on multiple trials (different team, different experimental setup)."

Anonymization

We acknowledge that for some of the works fitting this track, it may be difficult to submit in completely anonymized form, e.g., when links to demos, data, or software are required to assess the suitability of the submission for GECCO. Whenever possible, we strongly encourage the authors to make use of anonymous repositories (available on Zenodo and for GitHub repositories, for example). In the ideal case, these repositories will be deanonymized only after the notification. Where it is impossible to anonymize repositories, the BBSR track allows to link resources that possibly reveal authors’ identity. However, also in this case, all other elements of the paper shall follow the standard anonymization guidelines. In particular, we require that author names, affiliations, and acknowledgments are suppressed and that, to the maximum extent possible, references to any of the author's own work should be made as if the work belonged to someone else. We strongly recommend the use of the following option:

\documentclass[dvipsnames,format=sigconf,anonymous=true,review=true]{acmart}

CS - Complex Systems

Description

This track invites all papers addressing the challenges of scaling evolution up to real-life complexity. This includes both the real-life complexity of biological systems, such as artificial life, artificial immune systems, and generative and developmental systems (GDS); and the real-world complexity of physical systems, such as evolutionary robotics and evolvable hardware.

Artificial life, Artificial Immune Systems, and Generative and Developmental Systems all take inspiration from studying living systems. In each field, there are generally two main complementary goals: to better understand living systems and to use this understanding to build artificial systems with properties similar to those of living systems, such as behavior, adaptability, learning, developmental or generative processes, evolvability, active perception, communication, self-organization and cognition. The track welcomes both theoretical and application-oriented studies in the above fields. The track also welcomes models of problem-solving through (social) agent interaction, emergence of collective phenomena and models of the dynamics of ecological interactions in an evolutionary context.

Evolutionary Robotics and Evolvable Hardware study the evolution of controllers, morphologies, sensors, and communication protocols that can be used to build systems that provide robust, adaptive and scalable solutions to the complexities introduced by working in real-world, physical environments. The track welcomes contributions addressing problems from control to morphology, from single robot to collective adaptive systems. Approaches to incorporating human users into the evolutionary search process are also welcome. Contributions are expected to deal explicitly with Evolutionary Computation, with experiments either in simulation or with real robots.

ECOM - Evolutionary Combinatorial Optimization and Metaheuristics

Description

The ECOM track aims to provide a forum for the presentation and discussion of high-quality research on metaheuristics for combinatorial optimization problems. Challenging problems from a broad range of applications, including logistics, network design, bioinformatics, engineering and business have been tackled successfully with metaheuristic approaches. In many cases, the resulting algorithms represent the state-of-the-art for solving these problems. In addition to evolutionary algorithms, the class of metaheuristics includes prominent generic problem solving methods, such as tabu search, iterated local search, variable neighborhood search, memetic algorithms, simulated annealing, GRASP and ant colony optimization.

Scope

The ECOM track encourages original submissions on the application of evolutionary algorithms and metaheuristics to combinational optimization problems. The topics for ECOM include, but are not limited to:

• Representation techniques
• Neighborhoods and efficient algorithms for searching them
• Variation operators for stochastic search methods
• Search space and landscape analysis
• Comparisons between different techniques (including exact methods)
• Constraint-handling techniques
• Automated design of combinatorial optimisation algorithms
• Characteristics of problems and problem instances

Notice that the submission of very narrowed case studies of real-life problems as well as highly specific theoretical results on the performance of evolutionary algorithms may be better suited to other tracks at GECCO.

EML - Evolutionary Machine Learning

Description

The Evolutionary Machine Learning (EML) track at GECCO covers advances in the theory and application of using evolutionary computation methods to solve Machine Learning (ML) problems. Evolutionary methods can tackle many different tasks within the ML context, including problems related to supervised, unsupervised, semi-supervised, and reinforcement learning, as well as more recent topics such as transfer learning and domain adaptation, deep learning, representation learning, interpretability of machine learning models, and learning with unbalanced data and missing data.

The global search capability featured by evolutionary methods provides a valuable complement to the local search process that typically underpins non-evolutionary ML methods, and combinations of the two often demonstrate desirable promise in practice.

We encourage submissions related to theoretical advances, innovation of new algorithms, and renovation/improvement of existing algorithms, as well as application-focused papers. Authors are strongly encouraged to compare their EML approaches to the corresponding state-of-the-art non-evolutionary ML methods, where appropriate.

If your work focuses on the use of ML for solving evolutionary computation problems, please consider the new L4EC track that complements this track.

Scope

More concretely, topics of interest include but are not limited to:

• Theoretical and methodological advances on EML
• Evolutionary ensemble learning
• Evolutionary transfer learning and domain adaptation
• Evolutionary transfer learning and domain adaptation
• Evolutionary representation learning
• Learning Classifier Systems (LCS) and evolutionary rule-based systems
• Evolutionary computation techniques (e.g. genetic programming, particle swarm optimisation, and differential evolution) for solving ML tasks such as clustering, dimension reduction (feature selection, extraction, and construction), and representation learning
• AutoML (e.g. hyper-parameter tuning for ML) via evolutionary methods
• EML with a small number of examples, unbalanced data or missing data
• Visualizing or improving the interpretability of ML models via evolutionary approaches
• Parallel, distributed, and decentralized EML, including approaches based on high performance computing (with GPUs/TPUs), cloud computing ,and edge computing as well as federated learning
• Applications of EML (non-exhaustive list):
  • Computer vision and image processing
  • Pattern recognition and data mining
  • Bioinformatics, life sciences, medicine, and health
  • Space technology
  • Cognitive systems and modelling
  • Economic modelling
  • Intelligent transportation
  • Cyber security

EMO - Evolutionary Multiobjective Optimization

Description

In many real-world applications, multiple objective functions must be optimized simultaneously, leading to a multiobjective optimization problem (MOP), for which a single ideal solution seldomly exists. Instead, MOPs typically admit multiple compromise solutions representing different trade-offs among the objectives. Due to their applicability to a wide range of MOPs, including black-box problems, evolutionary and other randomized optimization algorithms for multiobjective optimization have given rise to an important and active research area known as Evolutionary Multiobjective Optimization (EMO). EMO algorithms do not require continuity or differentiability assumptions and can handle problem characteristics such as nonlinearity, multimodality, and stochasticity. Furthermore, preference information from a decision-maker can be used to deliver a finite-size approximation to the optimal solution set (the Pareto-optimal set) in a single optimization run.

Scope

The Evolutionary Multiobjective Optimization (EMO) Track brings together researchers from this field and related areas to explore all facets of EMO development and application. The track covers a wide range of topics, including but not limited to:

• Handling continuous, combinatorial, or mixed-integer MOPs
• Benchmarking methodologies, including test problems and performance assessment
• Benchmarking studies, especially in comparison to non-EMO approaches
• Selection and variation mechanisms
• Hybridization techniques, parallel and distributed models
• Software development and implementation aspects
• Convergence assessment and stopping criteria
• Theoretical foundations and search space analysis that bring new insights into EMO
• Visualization techniques for solution sets and multiobjective landscapes
• EMO algorithm selection and configuration
• Preference articulation and interactive EMO
• Many-objective optimization, large-scale optimization
• Computationally expensive objectives
• Constraint handling, uncertainty handling
• Real-world applications that go beyond solving specific problems, bringing new and broader insights into EMO

ENUM - Evolutionary Numerical Optimization

Description

The ENUM track (Evolutionary NUMerical optimization) is concerned with randomized search algorithms and continuous search spaces. The scope of the ENUM track includes, but is not limited to, stochastic methods such as differential evolution (DE), evolution strategies (ES), estimation-of-distribution algorithms (EDAs) and particle swarm optimization (PSO). The track is also concerned with the analyses of continuous search spaces to better understand the complexity of optimization problems and benchmarking of continuous optimization.

Scope

The ENUM track invites submissions that present original work regarding theoretical analysis, algorithmic design, and experimental validation of algorithms for optimization in continuous domains, including work on large-scale and budgeted optimization, handling of constraints, multi-modality, noise, uncertain and/or changing environments, and mixed-integer problems. Work that advances experimental methodology and benchmarking, problem and search space analysis is also encouraged.

Application papers reporting on solving a particular real-world optimization problem with continuous search space, with a relevant methodology, should be sent primarily to the Real-World Applications (RWA) track, with ENUM being a possible secondary track. On the other hand, if one or more "real-world-like" problems are used as a testbed for a comparison of several relevant methods, ENUM is the right primary track.

Papers dealing with theoretical analyses of evolutionary algorithms in continuous search spaces can be submitted primarily to the ENUM track with the theory track as a secondary track, or the other way round.

GA - Genetic Algorithms

Description

The Genetic Algorithm (GA) track has always been a large and important track at GECCO. We invite submissions to the GA track that present original work on all aspects of genetic algorithms, including, but not limited to:

• Practical, methodological and foundational aspects of GAs
• Design of new GA operators including representations, fitness functions, initialization, termination, parent selection, replacement strategies, recombination, and mutation
• Design of new and improved GAs
• Fitness landscape analysis
• Comparisons with other methods (e.g., empirical performance analysis)
• Design of tailored GAs for new application areas
• Handling uncertainty (e.g., dynamic and stochastic problems, robustness)
• Metamodeling and surrogate assisted evolution
• Interactive GAs
• Co-evolutionary algorithms
• Parameter tuning and control (including adaptation and meta-GAs)
• Constraint Handling
• Diversity management (e.g., fitness sharing and crowding, automatic speciation, spatial models such as island/diffusion)
• Bilevel and multi-level optimization
• Ensemble based genetic algorithms
• Model-Based Genetic Algorithms

As a large and diverse track, the GA track will be an excellent opportunity to present and discuss your research/application with a wide variety of experts and participants of GECCO.

GECH - General Evolutionary Computation and Hybrids

Description

General Evolutionary Computation and Hybrids is a track focusing on how EAs are used as part of larger systems in synergy with other algorithms, including hybrid methods and other, more general combinations of EAs with other components. We also welcome high-quality contributions on a wide range of EA topics which do not fit exclusively into other GECCO tracks. We don’t consider hybrids based only on superficial metaphors (Sörensen, 2015) as on-topic for this track.

Scope

Areas of interest include the following - but the limit should be set by your creativity not ours:

• Combining EAs with mechanisms to control or coordinate a set of algorithms, such as hyper-heuristics (selective and generative);
• Combining EAs with constructive heuristics;
• Combining EAs with classical methods (linear and integer programming, dynamic programming, constraint programming, etc.);
• Combining EAs and traditional AI methods such as A-star, tree search, Monte Carlo tree search;
• EAs incorporating multi-fidelity and multi-resolution objective function evaluation techniques;
• Hybridising approaches such as EA+EA (e.g., meta-EA), EA+PSO, EA+ACO, EA+LS (memetic), EA+Fuzzy;
• EA+A-life including co-evolutionary methods, both competitive and co-operative;
• Search algorithms combining quantum and classical computation;
• EAs using special techniques for parallel and distributed computing, or high performance hardware such as GPUs;
• Hybrid EAs which use landscape analysis techniques as part of the search.

GP - Genetic Programming

Description

Genetic Programming is an evolutionary computation technique that automatically generates solutions/programs to solve a given problem. In GP, various representations have been used, such as tree structures, linear sequences of code, graphs and grammars. Provided that a suitable fitness function is devised, computer programs solving the given problem emerge, without the need for humans to explicitly program the computer. The GP track invites original contributions on all aspects of evolutionary generation of computer programs or other executable structures for specific tasks.

Scope

Advances in genetic programming include but are not limited to:

• Analysis: Information Theory, Complexity, Run-time, Visualization, Fitness Landscape, Generalisation, Domain adaptation
• Synthesis: Programs, Algorithms, Circuits, Systems
• Applications: Classification, Clustering, Control, Data mining, Big-Data analytics, Regression, Semi-supervised Learning, Policy search, Prediction, Continuous and Combinatorial Optimisation, Streaming Data, Design, Inductive Programming, Computer Vision, Feature Engineering and Feature Selection, Natural Language Processing
• Environments: Static, Dynamic, Interactive, Uncertain
• Operators: Replacement, Selection, Crossover, Mutation, Variation
• Performance: Surrogate functions, Multi-Objective, Coevolutionary, Human Competitive, Parameter Tuning
• Populations: Demes, Diversity, Niches
• Programs: Decomposition, Modularity, Semantics, Simplification, Software Improvement, Bug Repair, Software/Program Testing
• Programming Languages: Imperative, Declarative, Object-oriented, Functional
• Representations: Cartesian, Grammatical, Graphs, Linear, Rules, Trees, Geometric and Semantic
• Systems: Autonomous, Complex, Developmental, Gene Regulation, Parallel, Self-Organizing, Software

L4EC - Learning for Evolutionary Computation

Description

Learning for Evolutionary Computation (L4EC) is a relatively new track introduced in 2024 to recognize high-quality research that uses machine learning or statistical techniques and concepts to improve heuristics and algorithmic components in the field of evolutionary computation (EC).

Scope

This track focuses on heuristics, methods and concepts that leverage machine learning (including deep learning and reinforcement learning) or statistics to enhance EC methods. As such, topics of interest include, but are not limited to:

• Methods for automated algorithm design, selection, and configuration,
• Mechanisms that learn how to control or coordinate a set of EC algorithms, such as parameter tuning, parameter control, dynamic algorithm selection/configuration, and meta-heuristics,
• EC algorithms integrating methods to extract knowledge from the population dynamics, search trajectory and/or the genotype,
• Surrogate-based or surrogate-assisted optimization of expensive fitness functions, including multi-fidelity approaches,
• Feature-based methods that learn to characterize optimization problems, such as exploratory landscape analysis (ELA) and fitness landscape analysis.

In focusing on the use of learning methods for EC, this track complements the existing EML track, which focuses on the use of EC for machine learning problems.

NE - Neuroevolution

Description

Neuroevolution is a machine learning approach that applies evolutionary computation (EC) to constructing artificial neural networks (NNs). Compared with other neural network training methods, Neuroevolution is highly general and allows learning without explicit targets, with arbitrary neural models and network structures. Neuroevolution has been successfully used to address challenging tasks in a wide range of areas, such as reinforcement learning, supervised learning, unsupervised learning, image analysis, computer vision, and natural language processing.

The Neuroevolution track at GECCO aims to encourage knowledge exchange between interested researchers in this area. It covers advances in the theory and applications of Neuroevolution, including all different EC methods for evolving all types of neural networks, alone and in combination with other neural learning algorithms. Authors are invited to submit their original and unpublished work to this track.

Scope

More concretely, topics of interest include but are not limited to:

• Neuroevolution algorithms involving:
  • Any EC method, e.g. genetic algorithms, evolutionary strategy, and genetic programming, particle swarm optimisation, differential evolution, meta-heuristics, Quality-Diversity, and hybrid methods.
  • Any type of neural networks, e.g. Convolutional neural network (CNN), Recurrent neural network (RNN), Long short-term memory (LSTM), Deep Belief Network (DBN), transformers, and autoencoders.
  • Evolutionary neural architecture search
  • Optimisation of network hyperparameters, activation and loss functions, learning dynamics, data augmentation, and initialisation
  • Novel candidate representations
  • Novel search mechanisms
  • Novel fitness functions
  • Surrogate assisted Neuroevolution
  • Methods for improving efficiency
  • Methods for improving regularisation
  • Multi-objective Neuroevolution
  • Neuroevolution for reinforcement learning, supervised learning, unsupervised learning
  • Neuroevolution for transfer learning, one-short learning, few-short learning, multitask learning
  • Parallelised and distributed realisations of Neuroevolution
  • Combinations of Neuroevolution and other neural learning algorithms
  • Interpretable/explainable model learning
• Applications of Neuroevolution:
  • Computer vision, image processing and pattern recognition
  • Text mining, natural language processing
  • Speech recognition
  • Neural Architecture Search
  • Machine translation
  • Medical and biological problems
  • Evolutionary robotics
  • Artificial life
  • Time series analysis
  • Cyber security
  • Scheduling and combinatorial optimization
  • Healthcare
  • Finance, fraud detection and business
  • Social media data analysis
  • Game playing
  • Visualisation

RWA - Real World Applications

Description

The Real-World Applications (RWA) track welcomes rigorous experimental, computational and/or applied advances in evolutionary computation (EC) in any discipline devoted to the study of real-world problems. The RWA track covers also real-world problems arising in creative arts, including design, games, and music (having been merged with the former track DETA - Digital Entertainment Technologies and Arts), and search-based software engineering problems (the SBSE track having been discontinued). The aim is to bring together contributions from the diverse application domains into a single event.

Please note that, from 2025, the track will handle strictly new real-world applications. That is, applications that are newly modelled and solved with evolutionary algorithms. In other words, we will not consider articles that use previously existing, publicly available real-world benchmark problems. If you are using existing/published real-world benchmarks, you can submit to other GECCO tracks that are relevant to your algorithmic contributions.

The focus is on applications including but not limited to:

• Papers that present novel developments of EC, grounded in real-world problems.
• Papers that present new applications of EC to real-world problems.
• Papers that analyse the features of real-world problems, as a basis for designing EC solutions.
• Papers that would fall into the DETA domain, such as ones focussing on aesthetic measurement and control, biologically-inspired creativity, interactive environments and games, composition, synthesis and generative arts.
• Papers on search-based software engineering applications, such as automatic program repair, genetic improvement, software testing, requirements analysis, and project management.

All contributions should be original research papers demonstrating the relevance and applicability of EC within a real-world problem. Papers covering multiple disciplines are welcome; we encourage the authors of such papers to write and present them in a way that allows researchers from other fields to grasp the main results, techniques, and their potential applications. Papers on novel EC research problems and novel application domains of the arts, music, and games are especially encouraged.

Scope

The real-world applications track is open to all domains and all industries.

SI – Swarm Intelligence

Description

Swarm Intelligence (SI) is the collective problem-solving behavior of groups of animals or artificial agents that results from the local interactions of the individuals with each other and with their environment. SI systems rely on certain key principles such as decentralization, stigmergy, self-organization, local interaction, and emergent behaviors. Since these principles are observed in the organization of social insect colonies and other animal aggregates, such as bird flocks or fish schools, SI systems are typically inspired by these natural systems.
The two main application areas of SI have been optimization and robotics. In the first category, Ant Colony Optimization (ACO) and Particle Swarm Optimization (PSO) constitute two of the most popular SI optimization techniques with numerous applications in science and engineering, but other SI-based optimization algorithms are possible. Papers that study and compare SI mechanisms that underly these different SI approaches, both theoretically and experimentally, are welcome. In the second category, SI has been successfully used to control large numbers of robots in a decentralized way, which increases the flexibility, robustness, and fault-tolerance of the resulting systems.

Scope

The ACO-SI Track welcomes submissions of original and unpublished work in all experimental and theoretical aspects of SI, including (but not limited to) the following areas:

• Biological foundations
• Modeling and analysis of new approaches
• Hybrid schemes with other algorithms
• Multi-swarm and self-adaptive approaches
• Constraint-handling and penalty function approaches
• Combinations with local search techniques
• Approaches to solve multi- and many-objective optimization problems
• Approaches to solve dynamic and noisy optimization problems
• Approaches to multi-modal optimization, i.e., to find multiple solutions (niching)
• Benchmarking and new empirical results
• Parallel/distributed implementations and applications
• Large-scale applications
• Software and high-performance implementations
• Theoretical and experimental research in swarm robotics
• Theoretical and empirical analysis of SI approaches to gain a better understanding of SI algorithms and to inform on the development of new, more efficient approaches
• Position papers on future directions in SI research
• Applications to machine learning and data analytics

THEORY - Theory

Description

The theory track welcomes all papers performing theoretical analyses or concerning theoretical aspects in evolutionary computation and related areas. Results can be proven with mathematical rigor or obtained via a thorough experimental investigation.

In addition to traditional areas in evolutionary computation like Genetic and Evolutionary Algorithms, Evolutionary Strategies, and Genetic Programming we also highly welcome theoretical papers in Artificial Life, Ant Colony Optimization, Swarm Intelligence, Estimation of Distribution Algorithms, Generative and Developmental Systems, Evolutionary Machine Learning, Search Based Software Engineering, Population Genetics, and more.

Scope

Topics include (but are not limited to):

• analytical methods like drift analysis, fitness levels, Markov chains, large deviation bounds,
• dynamic and static parameter choices,
• fitness landscapes and problem difficulty,
• population dynamics,
• problem representation,
• runtime analysis, black-box complexity, and alternative performance measures,
• single- and multi-objective problems,
• statistical approaches,
• stochastic and dynamic environments,
• variation and selection operators.

Papers submitted to the theory track may contain an appendix to give additional information. The appendix will not be part of the proceedings, and is consulted only at the discretion of the program committee. All technical details necessary for a proper evaluation must be contained in the 8-page submission or in the appendix, including full proofs and/or complete descriptions of experiments.