Toward smarter wireless networks: EURECOM’s role in shaping semantic communications for 6G
The future of wireless communication promises to be faster, more reliable, and undeniably smarter. In the Horizon Europe 6G-GOALS (6G Goal-Oriented AI-enabled Learning and Semantic Communication Networks) project, EURECOM’s team led by Prof. Photios Stavrou and Prof. Marios Kountouris, and supported by PhD student Symeon Vaidanis and postdoctoral researcher Dr. Minjie Tang, along with their partners, is shaping a bold new vision: a communication paradigm that accounts for the importance of information in light of the users’ intentions and goals.
From data-centric to semantic-centric communications
Conventional networks are agnostic to the significance, importance, and value of the information they transmit. Whether the data is vital, redundant, or irrelevant, the network treats it the same. This traditional approach, although successful so far, is becoming increasingly inefficient in the era of agentic AI and hyperconnected intelligence, where enormous volumes of data are transmitted from distributed sensors, autonomous agents, and edge AI devices.
That’s where the 6G-GOALS project comes in and why EURECOM’s contribution is crucial. Our team is laying the theoretical and algorithmic foundations for semantic and goal-oriented communications, enabling networks to focus only on what matters most to the task at hand.
At its core, semantic communication is about identifying what is important and processing and transmitting only the right type and amount of information, in a timely manner, rather than raw data. It’s not necessary to send a whole image if the receiver only needs to know whether there’s a pedestrian in it for example. This shift has profound implications for communication efficiency, energy consumption, and even sustainability. EURECOM’s researchers explore how to quantify the “semantics of information”, its relevance, usefulness, importance, and goal-oriented utility and how to represent, compress, transmit, and reconstruct only what matters.
Mathematical foundations and topological insights
“Our approach begins by defining semantic information in a way that’s mathematically rigorous”, explains Prof. Stavrou. “In this context, EURECOM plays a central role in developing the theoretical foundations of semantic communications. In particular, the team leads Mathematical Definitions and Fundamental Limits, aiming to develop information-theoretic models that characterize how semantics can be quantified and transmitted in networked systems. This includes defining semantics-aware performance metrics that reflect not only the intrinsic value of information (e.g., structural features of an image) but also its utility for the communication goal — whether that’s recognizing an object, making a control decision, or triggering an alert.
In a nutshell, EURECOM researchers are developing generalizable models for multi-agent semantic networks. Their work treats sampling, representation, and compression in a unified, goal-driven framework. To make these models computationally practical, they are also creating new optimization algorithms using both convex and non-convex methods.
Approximating semantics and smarter information compression
Another major novelty of EURECOM’s work is exploring approximate methods that reduce the computational and communication load of semantic systems. Instead of insisting on exact reconstructions, the focus is on capturing only what’s semantically sufficient for the task. This involves designing strategies for approximate semantic data representation and interpretation, guided by information-theoretical concepts and principles. These tools help quantify trade-offs between the complexity of representations and their informativeness for downstream tasks — enabling more efficient, real-time operation in constrained environments.
Complementing the theoretical work, EURECOM also contributes to developing machine learning–enabled compression and representation techniques. These methods aim to preserve task-relevant semantics in compressed representations, using either reinforcement learning approaches or generative models. Unlike traditional compression, which focuses purely on fidelity, semantic compression targets relevance and perception — ensuring that what’s transmitted aligns with what the receiver actually needs to infer or decide. This approach is crucial for emerging applications where edge devices must collaborate efficiently under tight bandwidth and energy constraints, such as in cooperative robotics, intelligent sensing, and autonomous mobility.
Towards the future of communications and 6G
EURECOM’s contributions to the 6G-GOALS project are helping to lay the scientific groundwork for a more adaptive, energy-efficient, and intelligent communication and networking ecosystem. By rethinking what it means to “communicate” in a wireless network, the team is pushing the boundaries of what 6G can achieve while receiving only what is necessary. As networks evolve to support ever more autonomous and collaborative systems, the ability to transmit semantically important data will be central to their performance, sustainability, and societal value.
References
- Goal-Oriented and Semantic Communication in 6G AI-Native Networks: The 6G-GOALS Approach, 2024 European Conference on Networks and Communications & 6G Summit (EuCNC/6G Summit): Physical Layer and Fundamentals (PHY).
- This work has received funding from the Smart Networks and Services Joint Undertaking (SNS JU) under the European Union’s Horizon Europe research and innovation program under Grant Agreement No 101139232.
