Livia Elena Chatzieleftheriou (Technische Universiteit Delft)

Keynote: From Applied Mathematics to Automated Heuristic Design: Engineering Sustainable and Explainable 6G Networks.

Abstract: As we transition towards 6G, the integration of Integrated Sensing and Communications (ISAC) and Reconfigurable Intelligent Surfaces (RIS) promises a leap in network capability. However, this evolution faces a critical dual challenge: the increasing computational complexity of network orchestration and the urgent need for sustainability. Traditional AI approaches, while powerful, often function as energy-intensive “black boxes” that lack the transparency and efficiency required for large-scale, green deployments. In this keynote, we discuss two approaches for efficient, large-scale network deployment. We first introduce the RIXISAC project, which combines online learning, applied math, eXplainable AI (XAI), digital twins, and real prototypes, to study RIS as a green enabler for intelligent and explainable ISAC systems. We then transition to the emerging field of Automated Heuristic Design (AHD). By leveraging Large Language Models (LLMs) to automate the creation of lightweight heuristics, we can achieve state-of-the-art performance in tasks like 5G decoding and resource allocation with a fraction of the effort. We argue that the future of networking lies in light-weight and rigorous intelligence, and we conclude that our approach of connecting the power of AI with the foundational rigor of applied math will render 6G not only smarter but also fundamentally more sustainable.

Biography: Livia Elena Chatzieleftheriou (Member, IEEE) received the M.Sc. degree in applied mathematics and the Ph.D. degree in computer science. She is currently with TU Delft as a Marie Skłodowska-Curie Actions (MSCA) Postdoctoral Fellow (PF). Previously, she was a Juan de la Cierva awardee with the IMDEA Networks Institute, and a Lecturer with the University Carlos III of Madrid (UC3M). Her research focuses on bridging the gap between network intelligence and  mathematical rigor, with specific interests in online learning, optimization, and eXplainable AI (XAI) for next-generation mobile networks. Her current work within her MSCA PF project “RIXISAC” explores the synergy between Integrated Sensing and Communications (ISAC) and Reconfigurable Intelligent Surfaces (RIS).

Paolo Dini (Centre Tecnològic de Telecommunicacions de Catalunya)

Keynote: Decentralized Intelligence: towards sustainable machine learning

Abstract: The rapid growth of sensors, mobile services, and industrial applications powered by artificial intelligence (AI) is driving an unprecedented expansion of distributed data generation. Traditional centralized AI processing architectures are increasingly inadequate, as modern information sources operate at the network edge and produce massive data streams far from cloud data centers. Continuously transmitting this data to remote servers is inefficient, latency-prone, and energy-intensive.
Edge intelligence has emerged as a promising paradigm to address these limitations by enabling data processing directly at, or close to, its point of origin. This shift reduces communication overhead, latency, memory usage, and energy consumption, while also enhancing privacy. The inherently distributed nature of mobile networks makes them ideal candidates for hosting future large-scale edge computing infrastructures, potentially reducing reliance on energy-hungry cloud facilities.
However, high-accuracy AI models (e.g., deep learning architectures) remain computationally demanding and data-hungry. These requirements often exceed the capabilities of local edge devices. Achieving decentralized intelligence therefore calls for collaborative learning paradigms that support efficient training and inference across heterogeneous and resource-constrained environments, without compromising model performance or sustainability.
Recent research focuses on designing flexible and scalable architectures that enable fast convergence, robustness to failures and adversarial threats, and substantial gains in energy efficiency. Key enabling technologies include federated learning, semantics-aware processing, and advanced representation learning.
In this keynote, I will provide an overview of the current landscape of collaborative and sustainable AI, examine the interplay among foundational technological enablers, explore their applicability across diverse domains, and discuss the open challenges that must be addressed to realize an energy-efficient and low-footprint AI ecosystem.

Biography: Paolo Dini is Leading Researcher (R4) within the Centre Tecnologic de Telecomunicacions de Catalunya (CTTC) in Spain, where he coordinates the activities of the Sustainable Artificial Intelligence research unit. His research interests include distributed optimization and optimal control, machine learning, multi-agent systems, cyber-physical systems. His research activity is documented in more than 100 peer-reviewed scientific journal and international conference papers. He has been principal investigator in several EU initiatives in the field of energy-efficient learning and computing (e.g., MSCA Greenedge, MSCA SCAVENGE, EIC SONATA). He is member of ELLIS (European Laboratory for Learning and Intelligent Systems) and Associated Editor of the IEEE Open Journal of the Computer Society.