Deliverable D4.1 - Foundation of Reasoning through Semantic-empowered Communications: First results

The semantic communication frameworks generally assume that agents share a common ontological language and employ compatible models for semantic extraction and interpretation. Under this assumption, communication errors are typically attributed to lossy semantic compression or incomplete knowledge representation. However, when agents operate with heterogeneous languages or semantic models, communication becomes prone to semantic noise, potentially degrading the performance of communication strategies. Moreover, when communication is instrumental in training the extraction and interpretation models, such a mismatch can contribute to additional model divergence between transmitters and receivers. To mitigate such mismatches, the 6G-GOALS project lays the foundation for next-generation AI-native networks by ensuring efficient, meaningful, and goal-oriented information exchange.

This deliverable presents the foundational efforts of the 6G-GOALS project in addressing semantic mismatches in AI-driven, goal-oriented 6G communication networks. We apply the principles from domain adaptation theory to the mismatch problem between sender(s) and receiver(s) with distinct languages and semantic rules (for instance, when using semantic extraction and interpretation models trained for different domains). Robust JSCC schemes are considered to ensure resilience to model mismatches. Theoretical and algorithmic methods based on optimal transport theory are proposed to identify and implement the set of measurable transport planes over semantic spaces, which enable cost-effective recovery of meaning at the receiver, as intended by the sender. These approaches support the development of goal-oriented communication policies derived from general domain transferability principles. Furthermore, the problem of mismatch in knowledge or beliefs among agents is investigated as an optimal transport problem, drawing connections with domain adaptation, and the intrinsic relationships with rate-distortion theory, which may lead to theoretical guarantees on the existence and achievability of optimal plans, as well as establishing information-theoretic equivalences. Finally, approximate reasoning is explored as a means to relax semantic constraints, enabling flexible interpretation and extended expressiveness in logical representations. 

In this document, several advancements are proposed in semantic communication, adaptive encoding and error correction, paving the way for enhanced, reliable AI-driven semantic communications, and driving high-performance, sustainable and reliable 6G systems.