Federated Learning with Feature Reconstruction for Vector Quantization based Semantic Communication (2508.03248v1)

Abstract: Recent advancements in semantic communication have primarily focused on image transmission, where neural network (NN)-based joint source-channel coding (JSCC) modules play a central role. However, such systems often experience semantic communication errors due to mismatched knowledge bases between users and performance degradation from outdated models, necessitating regular model updates. To address these challenges in vector quantization (VQ)-based image semantic communication systems, we propose FedSFR, a novel federated learning (FL) framework that incorporates semantic feature reconstruction (FR). FedSFR introduces an FR step at the parameter server (PS) and allows a subset of clients to transmit compact feature vectors in lieu of sending full local model updates, thereby improving training stability and communication efficiency. To enable effective FR learning, we design a loss function tailored for VQ-based image semantic communication and demonstrate its validity as a surrogate for image reconstruction error. Additionally, we provide a rigorous convergence analysis and present a differentially private variant of FedSFR, along with formal privacy analysis. Experimental results on two benchmark datasets validate the superiority of FedSFR over existing baselines, especially in capacity-constrained settings, confirming both its effectiveness and robustness.