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

The 2023 ACM SIGEVO Outstanding Contribution Awardees

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 2023, two amazing members of our community received this recognition: Prof. Emma Hart, and Dr. Carlos A. Coello Coello. To celebrate these distinctions, Emma and Carlos kindly answered our questions reflecting on their contributions and views on evolutionary computation as well as their valuable advice to young researchers in the field.

Q2: Which are your most significant technical contributions to Evolutionary Computation (EC)?

I was privileged to be involved in the early days of the field of hyper-heuristics [1] which is still going strong. I am also proud of some of the more recent work around building systems that learn – in evolutionary robotics, and in continually improving optimisers [2] and feature-free selection [3]

Q3: What are the current open problems, or topics where you think there are opportunities for substantial contributions in our field?

There is a huge opportunity for research that combines EC and ML – we are already seeing this for example in the new tracks at GECCO on ML for Evo and Evo for ML. In particular, I think we need to start to think about systems that continually learn. This could mean developing optimisation algorithms that improve over time based on past experience, using knowledge extracted every time an algorithm is run. Optimisers also need to be adaptive in response to changing characteristics of the instances they solve over time so their performance doesn’t degrade. Despite all the hype, I’m also excited by some of the opportunities for EC methods in generative AI, particularly in design applications, for example robotics.

Q4: Do you think the current AI hype is an opportunity or a threat for EC?

Probably a bit of both. The general hype can put people off but I think in EC we have been doing ‘generative AI’ for a long time, for example using EC for design (robots, objects) and in art – we just didn’t use the label. I think EC can contribute a lot to the current interest in generative design, and we are already seeing that at GECCO.

Q5: How do you view the visibility of the EC community in the larger computer science community?

It’s still not ‘mainstream’ in CS but actually I see it gaining interesting in a lot of other fields – for example I’m currently working with engineering colleagues on using EC to design earthquake protection barriers and another group in designing buildings – these communities are very excited about it!]

Q6: What advice do you have for the younger generation of researchers in the field?

Meet people! Talk to people at conferences, both your peers and more senior people, and take opportunities to volunteer for things. So many of my collaborations (and friends!) have come from gradually building up a network of people from all across the world. Also don’t be afraid to try out new ideas or to gradually shift to new topics.

Q2: Which are your most significant technical contributions to Evolutionary Computation (EC)?

My main research area has been the design of multi-objective evolutionary algorithms. I was lucky enough to be an early researcher in an area which is called today “evolutionary multi-objective optimization” (EMO) and this allowed me to make a few contributions that have been considered relevant by my peers. For example, in my research group, we designed the first micro genetic algorithm (with a population of only 4 individuals) for multi-objective optimization [1]. We also proposed the first Pareto-based multi-objective artificial immune system [2] and one of the earliest Pareto-based multi-objective particle swarm optimizers [3]

Additionally, we have made relevant contributions to the development constraint-handling techniques that adopt multi-objective concepts [4] and provided the first proof of convergence of a multi-objective artificial immune system [5].

More recently, we also made relevant contributions to large-scale multi-objective optimization [6] and to the identification of the real sources of difficulty of problems having many (i.e., more than three) objectives [7].

Q3: What are the current open problems, or topics where you think there are opportunities for substantial contributions in our field?

Focusing on evolutionary multi-objective optimization, I believe that there are many open problems that are of great importance. For example, how to deal with very expensive objective functions in an effective and efficient way. Scalability (both in decision variable space and in objective function space) is another topic which remains relevant despite the significant amount of research that has been conducted in the last few years. The reason for that is that it still has many issues that have not been properly addressed. For example, we need to analyze the sort of diversity that we should expect in a very high-dimensional objective space if our multi-objective evolutionary algorithms use relatively small population sizes (of a few hundreds of individuals). Also, it is necessary to conduct more research on performance indicators, since we currently know only one which is fully Pareto compliant (the hypervolume, which is very costly when dealing with more than six objectives).
Other topics that I consider relevant are: parallel multi-objective evolutionary algorithms (research in this area is still very scarce), the potential use of coevolution in certain types of multi-objective problems, the proper incorporation of machine learning techniques that aim to deal in a more efficient way with problems that have a high degree of complexity (e.g., because of their high dimensionality). Finally, the theoretical foundations of several aspects of evolutionary multi-objective optimization (e.g., run-time analysis) is also a pending task.

Q4: Do you think the current AI hype is an opportunity or a threat for EC?

It’s an opportunity. Most of the AI hype has been caused by the use of deep neural networks, which are a computational intelligence technique. Let’s keep in mind that evolutionary computation is also a computational intelligence technique, and we can certainly benefit of this. Now that many people are paying attention to neural networks, we can try to get them interested in evolutionary computation as well. Evolutionary computation offers techniques that can nicely complement the use of neural networks. In fact, in recent years, there has been an increased interest in evolutionary machine learning (e.g., using evolutionary algorithms to evolve neural network architectures). However, I envision many more interactions between evolutionary algorithms and neural networks. For example, I believe that it’s worth exploring the use of hybrids between evolutionary algorithms and reinforcement learning and that it’s worth developing lightweight (in terms of computational cost) machine learning techniques that can help evolutionary algorithms to perform a more “informed” search so that they can be more efficient and effective while consuming less computational resources.

Q5: How do you view the visibility of the EC community in the larger computer science community?

It has increased, but unfortunately, it’s still relatively low. Today, most textbooks on artificial intelligence include evolutionary algorithms, but the presence of evolutionary computation in the top AI conferences is still rather limited and it mainly includes work focused on evolutionary machine learning. Something interesting is that evolutionary computation has gained more visibility within optimization (many textbooks on optimization now include evolutionary algorithms as a direct search method). However, an important portion of the traditional computer science community still pays very limited attention to evolutionary computation.

Q6: What advice do you have for the younger generation of researchers in the field?

My main advice for them is that they should try to be disruptive. Today, much of the research conducted in some areas (e.g., in evolutionary multi-objective optimization) consists of replacing some component of a well-known algorithm by another one that marginally improves its performance. In most cases, there is little justification for these proposals and the changes also reflect little novelty. It’s necessary to have fresh ideas, even if they are radical and controversial. That’s the only way in which a research area may remain active and in constant evolution.

References

[1] Coello Coello, C.A., Toscano Pulido, G. (2001). A Micro-Genetic Algorithm for Multiobjective Optimization. In: Zitzler, E., Thiele, L., Deb, K., Coello Coello, C.A., Corne, D. (eds) Evolutionary Multi-Criterion Optimization. EMO 2001. Lecture Notes in Computer Science, vol 1993. Springer, Berlin, Heidelberg. https://doi.org/10.1007/3-540-44719-9_9
[2] Coello, C.A.C., Cortés, N.C. Solving Multiobjective Optimization Problems Using an Artificial Immune System. Genet Program Evolvable Mach 6, 163–190 (2005). https://doi.org/10.1007/s10710-005-6164-x
[3] C. A. C. Coello, G. T. Pulido and M. S. Lechuga, “Handling multiple objectives with particle swarm optimization,” in IEEE Transactions on Evolutionary Computation, vol. 8, no. 3, pp. 256-279, June 2004, doi: 10.1109/TEVC.2004.826067. https://ieeexplore.ieee.org/document/1304847
[4] Carlos A. Coello Coello (2000) Constraint-Handling using an Evolutionary Multiobjective Optimization Technique, Civil Engineering and Environmental Systems, 17:4, 319-346, DOI: 10.1080/02630250008970288 https://www.tandfonline.com/doi/abs/10.1080/02630250008970288
[5] Villalobos-Arias, M., Coello, C.A.C., Hernández-Lerma, O. (2004). Convergence Analysis of a Multiobjective Artificial Immune System Algorithm. In: Artificial Immune Systems. ICARIS 2004. Lecture Notes in Computer Science, vol 3239. Springer, Berlin, Heidelberg. https://doi.org/10.1007/978-3-540-30220-9_19
[6] L. M. Antonio and C. A. C. Coello, “Use of cooperative coevolution for solving large scale multiobjective optimization problems,” 2013 IEEE Congress on Evolutionary Computation, Cancun, Mexico, 2013, pp. 2758-2765, doi: 10.1109/CEC.2013.6557903. https://ieeexplore.ieee.org/document/6557903
[7] O. Schutze, A. Lara and C. A. C. Coello, “On the Influence of the Number of Objectives on the Hardness of a Multiobjective Optimization Problem,” in IEEE Transactions on Evolutionary Computation, vol. 15, no. 4, pp. 444-455, Aug. 2011, doi: 10.1109/TEVC.2010.2064321. https://ieeexplore.ieee.org/document/5601759

---

Overview of GECCO 2023

By Sara Silva and Luís Paquete

GECCO 2023 would not have been possible without an enormous team of dedicated people: 37 organisers, supported by the business committee; 8 members of the local team, together with 24 local volunteers and 55 international student volunteers; the chairs of the 13 tracks, the 24 workshops, including the student workshop, and the 12 competitions. We also thank all our speakers, including the 3 keynotes (Riccardo Poli, Carla Gomes, Kenneth De Jong) and the presenters of the 32 tutorials; all the reviewers; the sponsors and supporters; the chairs of the summer school that took place two days before the conference; and many more. We thank everyone for contributing to the efforts of this community event.

GECCO 2023 was organized into these 13 main tracks: Ant Colony Optimization and Swarm Intelligence (ACO-SI), Complex Systems (CS), Evolutionary Combinatorial Optimization and Metaheuristics (ECOM), Evolutionary Machine Learning (EML), Evolutionary Multiobjective Optimization (EMO), Evolutionary Numerical Optimization (ENUM), Genetic Algorithms (GA), General Evolutionary Computation and Hybrids (GECH), Genetic Programming (GP), Neuroevolution (NE), Real World Applications (RWA), Search-Based Software Engineering (SBSE) and Theory.

Among all the tracks, GECCO 2023 received 519 full paper submissions and accepted 180 of them, which resulted in a 34.7% acceptance rate. GECCO 2023 also received 55 poster paper submissions. The total number of accepted posters was 187, of which 148 from full paper submissions and 39 from poster paper submissions. Among the accepted full papers, 26 were nominated for the Best Paper Award. For the first time at GECCO, we have acknowledged those reviewers that excelled over their peers, awarding them the Outstanding Reviewer certificate.

Figure 1 shows the number of submissions (light blue), the number of accepted full papers (yellow), the acceptance rate of each track (solid blue line) and the acceptance rate of the conference (dotted blue line). The tracks are sorted by the number of full paper submissions they received (decreasing from left to right). The EML track received the largest number of submissions. The acceptance rate was different in each track, with the largest deviations from the average being observed in the Theory and ACO-SI tracks (55% and 42%, respectively) and in the GECH and ENUM tracks (20% and 27%, respectively).

The submissions received by the main tracks at GECCO 2023 were authored by 1460 researchers from 57 countries, with 80% of these authors affiliated with the 16 countries shown in Figure 2 (where the top axis shows the number of submissions). The USA was, once again, the country with most authors (170); also in the top 5 were Germany (147), China (137), United Kingdom (121), and France (103).

Besides the papers and posters of the main conference, GECCO 2023 also accepted 119 workshop papers (12 of which in the student workshop), 32 tutorials, 18 Late-Breaking Abstracts and 29 Hot-Off-the-Press contributions. The Humies event presented its 8 finalists, this year in a plenary session. These and other events were packed in four and a half amazing days of scientific and social program!

The entire organization team (some of us in Figure 5) thanks all the participants, onsite and online. We did our best to provide you with a fantastic conference, and it warms our hearts that the final GECCO survey earned us nothing less than 4.6 stars! We appreciate all your comments and suggestions for improvement. GECCO would not exist without you!

---

Software

We are thankful for the opportunity to announce our Python package “moptipy,” which offers a rich set of tools for implementing, applying, and experimenting with metaheuristic optimization algorithms. Our package has been designed with benchmarking in mind and thus provides very comprehensive experimentation, data collection, and evaluation facilities out of the box. moptipy has the following features:

1. Very comprehensive documentation with many examples, reaching down to literature references inside the code and up to complex example experiments.
2. Several standard search spaces (bit strings, permutations, real vectors), operators, and algorithms, including randomized local search/hill climbers, simulated annealing, evolutionary algorithms, memetic algorithms, NSGA-II, several numerical optimization algorithms, etc., are already implemented and ready for use.
3. It is very easy to implement new algorithms using moptipy, be they general black-box methods or tailored for specific optimization problems.
4. It is also easy to integrate algorithms from external libraries and unify them under our API, which we did as proof-of-concept with CMA-ES, BOBYQA, as well as for the algorithms from SciPy.
5. Data can be collected at different verbosity levels, ranging from only providing the final result and its quality via the API (without creating any files) to creating log files with all (or all improving) steps of an algorithm, the result, the algorithm and problem parameters, the system setup, non-dominated solutions, and the random seed for fully-reproducible experiments. [1]
6. An experiment execution facility for very simple and robust parallel and distributed experimentation is provided. Parallelization and distribution works out-of-the-box based on shared folders and thus does not require additional libraries or programming effort [1].
7. Stopping criteria for optimization processes can be defined based on goal solution qualities, clock time, and/or consumed objective function evaluations.
8. All experiments are fully reproducible, i.e., from a log file an algorithm and problem can be configured such that, in the replication experiment, exactly the same search steps are performed as in the original setting [1].
9. The experiment evaluation facility can parse the log files and generate progress plots, result tables, ERT and ECDF plots, statistical test tables, and export data towards Excel or the popular IOHanalyzer.
10. Both single-objective and multi-objective optimization are supported under the same unified API.
11. The package has a good unit test coverage and pre-defined tools to unit test your own code. When implementing new objective functions, algorithms, encodings, oder operators, it is possible to use these tools to look for errors.

The package is written in Python (>= 3.10), which currently probably is the predominant language in machine learning and AI as well as maybe the most-often used language in university classes. moptipy is therefore ideal for the use by both students and practitioners in AI, ML, or computer science in general. It can be installed from PyPI simply by doing

pip install moptipy

After the installation, the many examples listed at https://thomasweise.github.io/moptipy/#examples can be executed to get familiar with the package.

Some of the design rationales behind moptipy are discussed in

[1]. Thomas Weise and Zhize Wu. Replicable Self-Documenting Experiments with Arbitrary Search Spaces and Algorithms. In Sara Silva and Luís Paquete, eds., Companion Proceedings of the Conference on Genetic and Evolutionary Computation (GECCO’23), Lisbon, Portugal, July 15-19, 2023, pages 1891-1899. ACM. 2023. doi:10.1145/3583133.3596306.

---

Announcements

• Editorial to the “Evolutionary Reinforcement Learning” Special Issue
  Adam Gaier, Giuseppe Paolo, Antoine Cully, Article No.: 9, pp 1–2
  https://doi.org/10.1145/3624559
• Combining Evolution and Deep Reinforcement Learning for Policy Search: A Survey
  Olivier Sigaud, Article No.: 10, pp 1–20
  https://doi.org/10.1145/3569096
• P2P Energy Trading through Prospect Theory, Differential Evolution, and Reinforcement Learning
  Ashutosh Timilsina, Simone Silvestri, Article No.: 11, pp 1–22
  https://doi.org/10.1145/3603148
• Curiosity Creates Diversity in Policy Search
  Paul-Antoine Le Tolguenec,Emmanuel Rachelson,Yann Besse,Dennis G. Wilson
  Article No.: 12, pp 1–2
  https://doi.org/10.1145/3605782

---

---

Forthcoming Events

Call for Papers

Important Dates

---

GI @ ICSE 2024 Genetic Improvement Workshop

The 13th instalment of the GI workshop will take place in Lisbon, collocated with the International Conference on Software Engineering, ICSE 2024.

Authors of selected accepted papers will be invited to submit an extended version of their papers to an ASE special issue on Genetic Improvement.

Important Dates

• Submission Deadline:  7 December 2023 (Thu)
• Notification:  11 January 2023 (Thu)
• Camera-ready: 25 January 2024 (Thu)
• Workshop: 16 April 2024 (Tue)

---

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 31, 2024: 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 14, 2024: Finalists will be notified by e-mail
• Friday, June 28, 2024: Finalists who will not be in Melbourne to present in person must
  submit a 10-minute video presentation to goodman at msu dot edu. Finalists who will present in person must submit a copy of their slides, for the advance use of the judges, to goodman at msu dot edu.
• July 14-18, 2024 (Sunday – Thursday): 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).
• Thursday, July 18, 2024: Announcement of awards at the plenary session of the GECCO conference

---

About this Newsletter