Goal-oriented Communications based on Recursive Early Exit Neural Networks
Jary Pomponi, Mattia Merluzzi, Alessio Devoto, Mateus Pontes Mota, Paolo Di Lorenzo, Simone Scardapane
TL;DR
The paper tackles goal-oriented semantic communications over wireless links where devices face limited compute and energy resources. It introduces Recursive Early Exit Neural Networks (REEN) with layer-wise auxiliary exits and a recursive fusion mechanism, enabling early classification and dynamic partitioning across device and edge. An online reinforcement-learning optimizer jointly selects exit points, splitting decisions, and offloading under time-varying channel conditions, using margin-based proxies and end-to-end delay constraints to guide decisions, with $D_{\text{tot}}$ and $m$ as key metrics. Numerical validation on edge inference tasks shows favorable latency, accuracy, and resource trade-offs, highlighting the practical potential of combining early exits with RL-based resource management for robust, scalable edge AI in networked environments.
Abstract
This paper presents a novel framework for goal-oriented semantic communications leveraging recursive early exit models. The proposed approach is built on two key components. First, we introduce an innovative early exit strategy that dynamically partitions computations, enabling samples to be offloaded to a server based on layer-wise recursive prediction dynamics that detect samples for which the confidence is not increasing fast enough over layers. Second, we develop a Reinforcement Learning-based online optimization framework that jointly determines early exit points, computation splitting, and offloading strategies, while accounting for wireless conditions, inference accuracy, and resource costs. Numerical evaluations in an edge inference scenario demonstrate the method's adaptability and effectiveness in striking an excellent trade-off between performance, latency, and resource efficiency.