Implicit-Reasoning-Enabled Elastic Semantic Communications for AI-Agent Networks
Organizations: National Key Laboratory of Wireless Communications, University of Electronic Science and Technology of China, Chengdu 611731, China · Tianfu Jiangxi Laboratory, Chengdu, China · School of Information and Communications Engineering, Xi’an Jiaotong University, Xi’an, Shaanxi 710049, China · Department of Communications and Computer Engineering, University of Malta, MSD 2080 Msida, Malta
Abstract
AI-agent networks introduce new communication demands driven by proactive interactions and autonomous task execution. Existing adaptive semantic communication schemes primarily adjust semantic representations and resource configurations within predefined processing scales and capability sets, which may be insufficient for evolving task requirements. To address this gap, we introduce elastic semantic communications (ESC), a task-driven framework that extends semantic adaptation to the runtime execution and participation of task-supporting capabilities. ESC comprises two complementary dimensions: vertical elasticity adjusts the execution intensity of existing capabilities, while horizontal elasticity reconfigures the participating capability set. The framework connects these dimensions through the expected marginal value of additional information and computation. To realize ESC, we propose an implicit-reasoning-enabled device--edge mechanism that supports variable-depth local inference through shared-parameter state updates and edge-side continuation through wireless transfer of intermediate reasoning states. Experiments on Sudoku-Extreme demonstrate that the benefits of additional inference depend on wireless conditions, while task-aware edge invocation can preserve task accuracy with substantially lower state-transfer overhead.
Figures & tables
Semantic Communication
Semantic communication sends meaning- or task-relevant information so receivers can act without reconstructing every transmitted bit or signal.
See all 117 papers