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

PhD Dissertation Reports

Towards Bio-Inspired Interpretable Embodied Artificial Intelligence

Giorgia Nadizar, University of Trieste, Italy

Supervisors: Eric Medvet, University of Trieste, Italy and Stefano Nichele, Østfold University College, Norway

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.

In addition, AI-controlled robots raise concerns related to trust and transparency, particularly due to the inherent difficulty in understanding the underlying decision-making processes. In fact, people tend to trust AI systems more when they are transparent and interpretable. Yet, many AI models applied in robotic control lack the necessary transparency, which makes them harder to trust.

This work focuses on these two interrelated aspects of embodied AI: the enhancement of biological resemblance in AI-controlled robots and the improvement of their interpretability. The goal is to advance their performance, gain a better understanding of their behavior, and ultimately move toward deployment in real-world settings.

Bio-Inspiration in Embodied AI

First, we studied how the integration of biological principles into AI for robotics can enhance the performance and autonomy of embodied agents, focusing also on to what extent in-silico findings reflect and illuminate phenomena in the biological domain.  We conducted our experiments on modular soft robots (MSRs), which serve as an ideal test bed due to their inherent similarity to biological organisms.

First, we experimented evolving different models of artificial neural networks (ANNs), ranging from classical multilayer perceptrons (MLPs) to more bio-inspired spiking neural networks (SNNs), showing how SNNs not only outperform MLPs in task effectiveness, but also enable more adaptive and robust behaviors.

We further explored neural plasticity, the mechanism behind brain adaptation and specialization in biological organisms. Our in-silico findings in the context of body-brain evolution showed strong analogy with biology, with plastic controllers showing clear evidence of specialization dictated by their experience during the robot lifetime. In addition, plastic controllers were significantly more efficient and effective.

Last, we introduced the biological principle of synaptic pruning, the removal of unnecessary brain synapses, to MSRs to reduce over-parameterization and work towards more compact controllers. We found that thanks to evolution even after removing a large fraction of synapses, up to 75%, the controllers could maintain full functionality, mirroring the behavior of biological organisms, whose brains naturally undergo synaptic pruning after an initial phase of excessive synapse formation.

Interpretability in Embodied AI

Next, we investigated how interpretability can be achieved in the control of embodied agents without sacrificing performance. For this part, we focused on rigid robots with simpler body dynamics to better isolate the influence of the controller.

We leveraged graph optimization techniques, namely Cartesian Genetic Programming and Linear Genetic Programming, to find graph control policies that are both interpretable and effective.

Our results showed that in simpler environments both GP techniques perform comparably to the state-of-the-art. In addition, the obtained controllers are significantly smaller and transparent, being composed, on average, of fewer than 20 operations.

However, in more complex problems with larger input and output spaces, we noted that GP struggled to achieve good performance, often converging prematurely to local optima that were difficult to escape. To address this problem, we relied on quality-diversity (QD) optimization, focusing on behavioral and graph diversity, to promote both performance and interpretability. Our findings showed how QD notably enhanced the performance of GP, yielding to better solutions across various settings.

Integrating Bio-Inspiration and Interpretability in Embodied AI

In the last part of this thesis, we focused on identifying a paradigm of bio-inspired interpretable embodied AI that can balance the complexity and the benefits of bio-mimicry with the need for transparency in control.

To this end, we experimented with graph-based controllers on MSRs, focusing on how QD can foster the optimization of MSRs that are interpretable, effective, and adaptable. We exploited QD to mimic bio-diversity, across the brain, the body, and the behavior axes. In our findings, we observed that diversity significantly enhanced performance and adaptability, especially with respect to body and behavior traits, underscoring their critical role in achieving robust, high-performing MSRs.

References

• Full list of publications
• Thesis manuscript

About the Author

---

Representing and Exploiting Benchmarking Data for Optimisation and Learning

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. 

Figure 1. The core entities in the OPTION ontology and their relations.

Building upon the semantically annotated benchmarking data, we conduct various empirical studies, including tasks such as algorithm performance prediction and automated algorithm selection (AAS). In the MLC domain, a data-driven AAS pipeline is proposed to exploit this MLC benchmarking data. We evaluate the predictive power of dataset meta-features for AAS and explore various ML approaches – including regression, classification, and pairwise methods – to identify the most effective one. In the BBO domain, we exploit benchmarking data about modular BBO algorithms to conduct a comprehensive analysis of how individual algorithm modules influence overall performance. We develop algorithm representations derived from performance and feature importance values, effectively linking algorithm behavior to problem landscape features. Using these representations, we also relate module configurations and performance, providing deeper insights into the impact of different modules on algorithm performance. Furthermore, the semantically annotated benchmarking data on modular BBO optimisation algorithms is used as a backbone for creating various knowledge graphs (KGs). The KGs are then examined for their predictive power in algorithm performance prediction. By applying scoring-based KG embedding methods and graph neural networks, we predict algorithm performance in transductive and inductive setups, respectively. Overall, the contributions of this dissertation include the development of ontology-based frameworks for managing benchmarking data in the ML and BBO domains, the creation of semantic data catalogs, and novel methodologies for algorithm selection and performance prediction. By addressing challenges in representation and exploitation, this work advances both ML and BBO. It provides tools for improved data management and algorithm selection, as well as insights into algorithm behavior.

Here is a link to the full dissertation: http://dx.doi.org/10.13140/RG.2.2.17860.54401

Related publications

A. Kostovska, J. Bogatinovski, S. Džeroski, D. Kocev, and P. Panov, “A catalogue with semantic annotations makes multilabel datasets FAIR”, Scientific Reports, vol. 12, no. 1, p. 7267, 2022.

A. Kostovska, D. Vermetten, C. Doerr, S. Džeroski, P. Panov, and T. Eftimov, “OPTION: OPTImization Algorithm Benchmarking ONtology”, IEEE Transactions on Evolutionary Computation, vol. 27, no. 6, pp. 1618–1632, 2023. doi: 10.1109/TEVC.2022.3232844.

T. Eftimov, G. Petelin, G. Cenikj, et al., “Less is more: Selecting the right benchmarking set of data for time series classification”, Expert Systems with Applications, vol. 198, p. 116-871, 2022.

A. Kostovska, D. Vermetten, P. Korošec, S. Džeroski, C. Doerr, and T. Eftimov, “Using machine learning methods to assess module performance contribution in modular optimization frameworks”, Evolutionary Computation, pp. 1–27, Aug. 2024, issn: 1063-6560. doi: 10.1162/evco_a_00356.

A. Kostovska, S. Džeroski, and P. Panov, “Semantic description of data mining datasets: An ontology-based annotation schema”, in Proceedings of International Conference on Discovery Science, Springer, 2020, pp. 140–155.

A. Kostovska, D. Vermetten, S. Džeroski, P. Panov, T. Eftimov, and C. Doerr, “Using knowledge graphs for performance prediction of modular optimization algorithms”, in International Conference on the Applications of Evolutionary Computation (Part of EvoStar), Springer, 2023, pp. 253–268.

A. Kostovska, A. Jankovic, D. Vermetten, et al., “Per-run algorithm selection with warmstarting using trajectory-based features”, in Proc. of Parallel Problem Solving from Nature (PPSN), Springer, 2022, pp. 46–60.

A. Kostovska, D. Vermetten, S. Džeroski, C. Doerr, P. Korosec, and T. Eftimov, “The importance of landscape features for performance prediction of modular CMA-ES variants”, in Proc. of Genetic and Evolutionary Computation Conference (GECCO), ACM, 2022, pp. 648–656.

A. Kostovska, D. Vermetten, C. Doerr, S. Džeroski, P. Panov, and T. Eftimov, “OPTION: OPTImization Algorithm Benchmarking ONtology”, in Proceedings of the Genetic and Evolutionary Computation Conference Companion, 2021, pp. 239–240.

A. Kostovska, C. Doerr, S. Džeroski, D. Kocev, P. Panov, and T. Eftimov, “Explainable Model-specific Algorithm Selection for Multi-Label Classification”, in 2022 IEEE Symposium Series on Computational Intelligence (SSCI), IEEE, 2022, pp. 39–46.

A. Jankovic, D. Vermetten, A. Kostovska, J. de Nobel, T. Eftimov, and C. Doerr, “Trajectory-based algorithm selection with warm-starting”, in 2022 IEEE Congress on Evolutionary Computation (CEC), IEEE, 2022, pp. 1–8.

A. Kostovska, G. Cenikj, D. Vermetten, et al., “PS-AAS: Portfolio Selection for Automated Algorithm Selection in Black-Box Optimization”, in International Conference on Automated Machine Learning, PMLR, 2023, pp. 11–1.

A. Kostovska, A. Jankovic, D. Vermetten, S. Džeroski, T. Eftimov, and C. Doerr, “Comparing Algorithm Selection Approaches on Black-Box Optimization Problems”, in Proceedings of the Companion Conference on Genetic and Evolutionary Computation, 2023, pp. 495–498.

A. Nikolikj, A. Kostovska, D. Vermetten, C. Doerr, and T. Eftimov, “Quantifying individual and joint module impact in modular optimization frameworks”, in 2024 IEEE Congress on Evolutionary Computation (CEC) (pp. 1-8). IEEE.

About the Author

Position Article

Designing Intelligence with Intelligence: The Future of AI and EC Collaboration

Yong-Hyuk Kim
School of Software, Kwangwoon University, Seoul, Republic of Korea
yhdfly@kw.ac.kr

Introduction: Designing Intelligence with Intelligence

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 rise of large-scale models like Transformers and large language models (LLMs) has deepened this interaction. AI now aids evolutionary design by predicting fitness, proposing mutation strategies, and learning optimization behaviors. Meanwhile, EC offers global search and model diversity, critical in complex tasks where local optimization is inadequate.

This article examines the deepening integration of these fields. We first review how EC has supported AI, then explore how AI is increasingly shaping EC. We conclude by highlighting key challenges and future directions in this mutual evolution. This shift goes beyond tool usage; it represents a mutual adaptation between two computational paradigms.

When Evolution Designs AI: How EC Has Enabled Modern AI

Evolutionary Computation (EC) has played a vital role in shaping modern AI systems. It has been used to optimize components that are difficult to fine-tune through conventional methods, including hyperparameters, model architectures, and symbolic structures.

One well-established application of EC is hyperparameter tuning. Evolutionary algorithms are particularly effective at exploring complex, non-differentiable parameter spaces, adjusting values such as learning rate, batch size, and dropout ratio. A recent example is OptFormer [Wang et al., 2020], which trains a Transformer model on past optimization data, much of which was generated using EC, to predict strong hyperparameter settings for new tasks.

EC has also contributed significantly to neural architecture search. For instance, the Evolved Transformer [So et al., 2019] used a genetic search strategy to outperform manually designed Transformer variants in natural language tasks. Similarly, HAT (Hardware-Aware Transformers) [Chen et al., 2022] used evolutionary search to identify latency-efficient Transformer architectures from a large design space, making them suitable for real-world deployment.

In domains where interpretability is critical, EC continues to prove useful. By integrating evolutionary strategies with Transformer-based models, researchers have built systems that can rediscover symbolic expressions from data. Jha et al. (2024) integrated pretrained Transformers with genetic programming, using the former to generate symbolic expressions and the latter to refine them, thereby enhancing both accuracy and robustness in symbolic regression.

When AI Designs Evolution: How Modern AI Models Assist EC

While EC has long been used to optimize AI systems, recent developments reveal a reverse trend: AI models, particularly Transformers and large language models (LLMs), are now being applied to enhance evolutionary computation itself. One notable example is the Evolution Transformer [Lange et al., 2024], which uses in-context learning to perform optimization by recognizing patterns from previous tasks. This eliminates the need for hand-crafted mutation or selection rules. These models do not replace evolution but are beginning to reshape how we implement and adapt evolutionary strategies in practice.

Pretrained models like LLMs are also being used to guide evolutionary search. These models can suggest candidate solutions from textual prompts, generate variation operators, and highlight promising regions of the search space (e.g., [Kim and Kim, 2025]). In prompt-driven workflows, they can even generate dynamic fitness functions or constraints on demand, significantly reducing manual effort. This capability is especially useful in domains like symbolic regression, program synthesis, and automated design. Additionally, hybrid systems are emerging where EC and AI components co-adapt. For example, EC may search for architectures while AI evaluates and refines them, forming a feedback loop. The Knowledge-Aware Evolutionary Transformer [Zhang et al., 2025] illustrates this synergy by evolving multitask architectures informed by domain knowledge. These developments signal a fundamental shift: AI is no longer merely optimized by EC; it now plays an active role in steering evolutionary dynamics.

Co-Evolving Intelligence: Toward a New Paradigm of AI-EC Integration

The collaboration between AI and EC is no longer one-sided. What began with EC optimizing AI, and later continued with AI assisting EC, has evolved into a deeper, reciprocal relationship. Each field contributes distinct strengths: EC is well-suited for noisy, multi-modal, and non-differentiable spaces, where it excels at global exploration and diversity preservation. AI models, particularly Transformers, offer strong generalization and can extract structure from complex data. When combined, EC generates diverse candidate solutions while AI contributes by interpreting results, guiding adjustments, and speeding up convergence.

This integration is becoming increasingly adaptive. EC can evolve the structure or behavior of AI systems, while AI components guide or enhance the evolutionary process in real time. The integration of learning and search offers a compelling route toward systems that adapt over time, but true self-improvement will depend on how well we manage complexity, constraints, and unintended behaviors. This trend outlines a future in which systems may progressively reshape their architecture and behavior through coupled processes of evolution and learning. In such a framework, AI and EC are no longer just tools; they become co-creators in artificial design.

Co-Designed Intelligence: Challenges and the Road Ahead

The convergence of Evolutionary Computation (EC) and Artificial Intelligence (AI) opens the door to fundamentally new approaches to intelligent system design. As these fields continue to integrate, they present significant opportunities but also raise practical and theoretical challenges that must be addressed for this collaboration to scale.

One major obstacle is computational cost. EC often involves evaluating thousands of candidate solutions, and large AI models already demand substantial resources. Without surrogate modeling, parallelization, or transfer learning, many hybrid approaches remain accessible only to well-funded research environments.

Another challenge is balancing generalization and specificity. EC is valued for its flexibility and domain-independence but may lack the fine-tuned precision required for specific tasks. In contrast, AI models often excel at narrowly defined problems but struggle to generalize. As architectures, loss functions, and optimization strategies grow more abstract and high-dimensional, EC will need additional guidance to navigate these spaces effectively.

Interpretability and control also pose concerns. In hybrid systems, the contribution of each component becomes harder to trace, which complicates reproducibility and transparency. These concerns are magnified in systems that evolve in unpredictable ways, sometimes beyond the understanding of their creators. Adding to these technical issues is a persistent interdisciplinary divide. The AI and EC communities still differ in terminology, evaluation criteria, and research cultures. Overcoming this gap will require shared benchmarks, collaborative venues, and ongoing exchange of ideas.

Despite these challenges, the integration of EC and AI offers a compelling path forward. Together, they can enable systems that adapt, learn, and refine themselves, pushing the boundaries of what intelligent design can achieve. For the SIGEVO community, this is not just a technical challenge but also a strategic opportunity: to lead the development of co-adaptive systems, to bridge conceptual divides, and to help define the future of intelligent computation with clarity and intent.

References

Chen, Y., Song, X., Lee, C., Wang, Z., Zhang, Q., Dohan, D., Kawakami, K., Kochanski, G., Doucet, A., Ranzato, M., Perel, S., & de Freitas, N. (2022). Towards learning universal hyperparameter optimizers with transformers. arXiv preprint arXiv:2205.13320.
 → (OptFormer)

So, D. R., Liang, C., & Le, Q. V. (2019). The Evolved Transformer. arXiv preprint arXiv:1901.11117.
 → (Evolved Transformer)

Wang, H., Wu, Z., Liu, Z., Cai, H., Zhu, L., Gan, C., & Han, S. (2020). HAT: Hardware-Aware Transformers for Efficient Natural Language Processing. arXiv preprint arXiv:2005.14187.
 → (HAT)

Jha, S., Gleyzer, S., Reinhardt, E., Baules, V., Charton, F., & Okada, N. (2024). Evolutionary and transformer-based methods for symbolic regression. In Proceedings of the NeurIPS 2024 Workshop on Machine Learning and the Physical Sciences.
 → (Symbolic Regression with Transformers)

Lange, R. T., Tian, Y., & Tang, Y. (2024). Evolution Transformer: In-context evolutionary optimization with transformers. arXiv preprint arXiv:2403.02985.
 → (Evolution Transformer)

Kim, T.-H. & Kim, Y.-H. (2025). Written by Artificial Intelligence, Evolved by Genetic Algorithms: An Evolutionary Challenge to Solving the Traveling Salesperson Problem with ChatGPT, GECCO 2025, accepted (to appear)

Zhang, R., Li, L., Jiao, L., & Yang, S. (2025). Knowledge-Aware Evolutionary Transformer. IEEE Transactions on Evolutionary Computation.
 → (Knowledge-Aware Evolutionary Transformer)

About the Author

---

Conference Overview

Summary of EvoStar 2025

Jéssica Parente (University of Coimbra) and Alina Geiger (University of Mainz)

The EvoStar conferences 2025 were held in Trieste, Italy, from 23 to 25 April. Trieste is a beautiful little city in the north-east of Italy, located directly by the sea. Some of us flew to Trieste and took the train from the airport to the city center. Along the way, many reconnected with old friends even before arriving in the city.

On the evening before the conference began, many of the PhD students gathered at the Student Reception to get to know each other over snacks and drinks. The highlight of the evening was a pub quiz, where the students could showcase their knowledge of Trieste, EvoStar, computer science, and Italian cuisine. Afterwards, the group headed into the city to enjoy a pizza dinner together.

The conference began the following day at the University of Trieste, which is located on a hilltop with a fantastic view over the Adriatic Sea and the surrounding mountains. After the official opening, Tea Tušar delivered an inspiring keynote on ideals and realities of benchmarking in evolutionary multiobjective optimization.

Afterwards, the first paper sessions began, featuring presentations on topics such as AI in Art and Music (EvoMUSART) or Computational Intelligence for Sustainability (EvoApps). The day ended with the poster session, where everyone had the opportunity to present their work, answer questions, and receive valuable feedback. As in previous years, the posters were creative and thoughtfully designed.

The next day began with the first EvoCOP session and several EvoApps presentations. This was followed by the Student Workshop, where the PhD students had the opportunity to engage with experienced researchers, learn about their career paths, the challenges they overcame during their PhDs, and how they manage to maintain a healthy work-life balance in academia. We would like to thank Darrell Whitley, Gabriela Ochoa, Gisele Pappa, Illya Bakurov, Sara Silva and Pablo García-Sánchez for sharing their experiences with us.

Student Workshop

Afterwards, we had lunch, followed by additional paper presentations across various tracks. At the end, we all gathered for a group photo to have a memory of this wonderful conference.

The social dinner kicked off at 7:30 PM at Trattoria Caprese, a lovely spot that brought together students and professors in a relaxed and friendly atmosphere. During the evening, Colin Johnson and Iñaki Hidalgo received the Evo* Award for Outstanding Contribution to Evolutionary Computation and were officially welcomed into the Old Crocs group! Also, Mário Giacobini received his award from last year.

Colin Johnson is a familiar face in the EvoStar community, thanks to his important contributions to both evolutionary art and genetic programming. For those who’ve published in EvoMUSART, Colin is a well-known name; he served several times as organizer and was part of the steering committee in the early days of the conference. Iñaki Hidalgo has been an important and active member of the SPECIES community since the very beginning. He was also one of the local chairs of EvoStar 2022, and he’s been a regular contributor by organizing several special sessions at EvoApplications. He is well known for his work applying evolutionary computation to biomedicine.

As usual, the night ended on a high note with the Evostar Choir, or in other words, everyone gathered to sing “Amigos para Siempre” together. Following the official closing of the dinner, those still in a good mood continued the evening at the local bars.

The last day started with a few more paper presentations and a very interesting talk by Daniela Besozzi on Mathematical modelling, computational intelligence and machine learning in biomedicine. Daniela demonstrated how fuzzy logic-based models can be effectively applied in biomedicine to analyze complex systems, using several compelling examples.

At the conference closing, the awards were presented. Among the many award winners, we would like to highlight:

• EvoAPPS best paper award: Emergent kin selection of altruistic feeding behaviour via non-episodic neuroevolution by Max Taylor-Davies, Gautier Hamon, Timothé Boulet and Clement Moulin-Frier
• EvoCOP best paper award: Evolutionary Anytime Algorithms by Aishwarya Prajna and Jonathan Rowe
• EuroGP best paper award: A Systematic Evaluation of Evolving Highly Nonlinear Boolean Functions in Odd Sizes by Claude Carlet, Marko Đurasević, Domagoj Jakobovic, Stjepan Picek and Luca Mariot
• EvoMUSART best paper award: Yin-Yang: Developing Motifs With Long-Term Structure And Controllability by Keshav Bhandari, Geraint A. Wiggins and Simon Colton
• Best Poster Award: Was Tournament Selection All We Ever Needed? A Critical Reflection on Lexicase Selection by Alina Geiger, Martin Briesch, Dominik Sobania and Franz Rothlauf

The Julian Francis Miller Award was presented to Jean-Baptiste Mouret for his pioneering work on the MAP-Elites algorithm that has exerted major influence beyond Evolutionary Computation to Machine Learning and Robotics. Further, a special award was presented to Marc Schoenauer this year for his many years of service and support: the Guardian Angel Award.

After the conference wrapped up, we all had lunch together, followed by the Species Society’s Annual General Meeting (AGM). It was a great chance for everyone to chat about the future of SPECIES and brainstorm new workgroups and projects to make our community even stronger. The Executive Board was re-elected, and we were excited to welcome Anna Kononova and Jamal Toutouh on board. On the flip side, we had to say goodbye to Iñaki Hidalgo and Kehinde Babaagba. Everyone shared a big thank you for all their hard work and dedication. They’ve been a huge part of what we’ve achieved, and we’re grateful for everything they’ve done.

After that, an optional guided city tour was offered in the afternoon for those interested in exploring more of Trieste’s rich history and culture. The tour began with a bus ride from the university to Piazza Sant’Antonio Nuovo, one of the city’s most iconic squares, known for its grand architecture and lively atmosphere.

One notable aspect of Trieste is its renowned coffee culture, especially as the home of the famous Illy brand. Often dubbed the coffee capital of Italy, the city offers an interesting curiosity: sipping espresso while standing is significantly cheaper than sitting down. This practice adds a social element to daily coffee breaks and keeps the experience both quick and affordable.  Besides, Trieste also holds an important place in European history. As a key economic hub during the Austro-Hungarian Empire, it served as a vital port linking Central Europe to the Mediterranean, a role that shaped much of its identity and legacy. We’d also like to give a heartfelt thanks to Alessandro Cesa for his helpful guidance throughout.

After the tour, unfortunately, it was already time to say goodbye. We really enjoyed our time in Trieste. Special thanks to the local chairs Eric Medvet, Luca Manzoni, Giorgia Nadizar and Gloria Pietropolli and the EvoStar coordinator Anna Esparcia-Alcázar for putting together such an amazing event.

We hope to see you all at Evostar 2026!

About the Authors

---

Announcements

New HorizonEU Project: AutoLearn-SI Launches at Jožef Stefan Institute!

Tome Eftimov, Jožef Stefan Institute, Slovenia

We’re excited to announce AutoLearn-SI, a €2.5M #HorizonEU-funded project at Jožef Stefan Institute (JSI), Slovenia, set to establish a world-class ERA Chair research group in automated machine learning and optimization (AutoML & AutoOPT).

AutoLearn-SI is on a mission to integrate next-generation AutoML and AutoOPT technologies into research, education, and innovation ecosystems, strengthening Slovenia’s position as a leader in AI-driven decision-making.

What makes AutoLearn-SI unique?

→ Led by an ERA Chair Holder from the USA, with collaboration from top research partners in France, Germany, the Netherlands, and Belgium.

→ Actively transferring knowledge to widening countries, including North Macedonia.

Why does it matter?

It’s a key step toward regional excellence in AI, machine learning, and optimization — empowering research, industry, and education across Europe. It represents a strategic step toward next-generation benchmarking by:

• Advancing automated benchmarking through AutoML & AutoOPT methods
• Focusing on experimental databases and algorithm selection/configuration — key pillars of trustworthy benchmarking
• Enabling dynamic, adaptive, and explainable benchmarking beyond static leaderboards
• Promoting open, FAIR benchmark resources and representation learning for landscape-aware evaluation
• Creating a living testbed for benchmarking research infrastructures within the European Research Area

Follow the journey & stay tuned for upcoming opportunities to collaborate!

---

Journal Latest Issues

---

---

---

Forthcoming Events

Keynotes

• Maria Amparo Alonso Betanzos, CITIC-University of A Coruña (UDC), Spain
• Javier Del Ser, University of the Basque Country (UPV/EHU), Spain
• Marc Schoenauer, Institut national de recherche en sciences et technologies du numérique (INRIA), France

---

The FOGA series aims at advancing our understanding of the working principles behind evolutionary algorithms and related randomized search heuristics, such as local search algorithms, differential evolution, ant colony optimization, particle swarm optimization, artificial immune systems, simulated annealing, and other Monte Carlo methods for search and optimization. Connections to related areas, such as Bayesian optimization and direct search, are of interest as well. FOGA is the premier event to discuss advances on the theoretical foundations of these algorithms, tools needed to analyze them, and different aspects of comparing algorithms’ performance

---

About this Newsletter