4th International Workshop on Artificial Intelligence for Autonomous computing Systems (AI4AS 2026)
Co-located with ACSOS 2026, which takes place in Cesena (Italy) - September 7-11, 2026.
Co-located with ACSOS 2026, which takes place in Cesena (Italy) - September 7-11, 2026.
All times in Anywhere on Earth (AoE) timezone.
| 9.20 - 9.30 | Opening |
| 9.30 - 9.50 | Adaptive Incremental Learning for Anomaly Detection under Domain Shift: An Additive Manufacturing Case Study. Akram Zarchini, Denis Dowling, Fatemeh Golpayegani |
| 9.50 - 10.10 | Discovering Collaboration from Novelty: Random Network Distillation for Clustered Federated Learning. Davide Domini, Gianluca Aguzzi, Ivana Dusparic, Danilo Pianini, Mirko Viroli |
| 10.10 - 10.30 | 3NLM: A Normative Logic Framework for Expectation-Driven Self-Adaptive Explanations. Zahra Atf, Nathan Lloyd, Peter Lewis |
| 10.30 - 10.50 | S3Eval: A Three-Layer Pipeline for Automated Evaluation of PDDL-Encoded Attack Paths. Lin Cui, Vincenzo Scotti, Raffaela Mirandola |
| Break | |
| 11.15 - 12.15 | Keynote: Sustainable Intelligence: Adaptive AI Systems for Environmental Efficiency in Autonomous and Distributed Computing, Monica Vitali |
| 12.15 - 12.35 | Towards Autonomous Surface Vessels: A World-Model-Centric Approach. Raphael Schwinger, Nils Bischoff, Mats Kurz, Lukas Nolte, Tim Nolte, Bahne Thiel-Peters, Sven Tomforde |
| 12.35 - 12.55 | Perception Risk for Path Planning in Autonomous Rover Navigation. Christian Medeiros Adriano, Mostafa Wael Aboalfotoh, Sona Ghahremani, Holger Giese |
| 12.55 - 13.00 | Closing |
Sustainable Intelligence: Adaptive AI Systems for Environmental Efficiency in Autonomous and Distributed Computing
Speaker: Prof. Monica Vitali (Politecnico di Milano, Italy)
Abstract. Artificial Intelligence is rapidly becoming the decision-making engine of autonomous and distributed systems, enabling increasingly capable applications while also raising important concerns about their environmental footprint. As AI models grow in complexity and scale, achieving sustainability requires moving beyond the optimization of individual algorithms toward a holistic view of AI systems, where computational resources are managed intelligently throughout the entire lifecycle. This keynote explores the vision of adaptive sustainable AI, where environmental impact is treated as a first-class design objective alongside performance, reliability, and quality of service. The central premise is that sustainability is not only a matter of developing more efficient models, but also of enabling AI systems to continuously decide what to compute, when to compute it, where to compute it, and at what level of fidelity according to application requirements and resource constraints. Through a set of research directions spanning different phases of the AI lifecycle, the talk highlights a common principle: environmental sustainability can be achieved by making AI systems adaptive in the way they acquire data, allocate computational resources, orchestrate distributed intelligence, and select inference strategies. This perspective opens new opportunities for designing autonomous systems that are not only intelligent in the services they provide, but also in the way they manage their own computational footprint, paving the way toward a new generation of environmentally aware AI.
About the speaker: Monica Vitali is Associate Professor at the Department of Electronics, Information and Bioengineering (DEIB) of Politecnico di Milano (Italy), where she collaborates with the Research on Advanced Information Systems Engineering (RAISE) group. Her research interests include adaptive information systems, data management in heterogeneous infrastructures, energy- and carbon-aware information systems management, and green data-centric AI. Since 2025, she is the chair of the Green ICT working group of Informatics Europe.
Modern computing systems are increasingly heterogeneous and operate at unprecedented scale across the cloud–edge continuum. Their structural and operational complexity often exceeds what can be effectively managed through manual configuration or static control logic, particularly when timely decisions are required in highly dynamic environments and strict Quality-of-Service (QoS) guarantees must be enforced.
The rapid evolution of Artificial Intelligence (AI) and Machine Learning (ML) techniques – including generative AI, agentic AI, edge intelligence, as well as collaborative and federated learning – has opened new avenues for engineering more robust, sustainable, and secure computing infrastructures. AI/ML techniques are increasingly embedded into the control loops of modern systems to enable self-adaptation. They are leveraged, for example, to extract actionable insights from high-dimensional and noisy monitoring data, to learn performance and workload models, to forecast internal or external dynamics (e.g., workload fluctuations or network conditions), and to automatically plan—and potentially enact—adaptation and resource management actions. Moreover, to scale this self-adaptive intelligence, collaborative learning paradigms enable multiple nodes or administrative domains to jointly build these predictive or prescriptive models by sharing knowledge rather than raw data, which is essential in large-scale distributed edge and cloud settings.
However, there are still several challenges to face for researchers and practitioners aiming to take advantage of these methodologies and incorporate them in their systems. Fundamental issues towards the applicability of AI and ML techniques across diverse domains must be investigated, especially as regards the accuracy, robustness, explainability, safety, security, performance and sustainability of AI-driven autonomous computing systems.
In this workshop, we solicit high-quality contributions that fit with the overarching theme of AI and ML meeting autonomous computing systems.
The aim of the workshop is to share new findings, exchange ideas and discuss research challenges on the following topics (not an exhaustive list):
All submissions are required to be formatted according to the standard IEEE Computer Society Press proceedings style guide. Papers can be submitted in PDF format via EasyChair, making sure to select the track “AI4AS-Workshop”. Submitted manuscripts must be no longer than 6 pages (including figures, tables, and references).
Accepted papers will be published in the ACSOS Companion volume and will appear in IEEE Xplore.
As per the standard IEEE policies, all submissions should be original, i.e., they should not have been previously published in any conference proceedings, book, or journal and should not currently be under review for another archival conference. We would like to also highlight IEEE’s policies regarding plagiarism and self-plagiarism, available here.
Moreover, as per IEEE guidelines, the use of content generated by artificial intelligence (AI) in a submission (including but not limited to text, figures, images, and code) shall be disclosed in the acknowledgments section.