Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Federated Learning Enhanced by Feature Reconstruction for Semantic Communication Module Updates of Agents

Yoon Huh, Bumjun Kim, Wan Choi

TL;DR

FedSFR is proposed, a novel federated learning framework that incorporates semantic feature reconstruction (FR) 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.

Abstract

Recent advancements in semantic communication have primarily focused on image transmission, where neural network-based joint source-channel coding modules play a central role. However, such systems often experience semantic communication errors due to mismatched knowledge bases between agents and performance degradation from outdated models, necessitating regular model updates.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 framework that incorporates semantic feature reconstruction (FR). FedSFR introduces an FR step at the parameter server 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. We further establish a rigorous convergence analysis of FedSFR. 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.

Federated Learning Enhanced by Feature Reconstruction for Semantic Communication Module Updates of Agents

TL;DR

FedSFR is proposed, a novel federated learning framework that incorporates semantic feature reconstruction (FR) 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.

Abstract

Recent advancements in semantic communication have primarily focused on image transmission, where neural network-based joint source-channel coding modules play a central role. However, such systems often experience semantic communication errors due to mismatched knowledge bases between agents 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 framework that incorporates semantic feature reconstruction (FR). FedSFR introduces an FR step at the parameter server 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. We further establish a rigorous convergence analysis of FedSFR. 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.

Paper Structure

This paper contains 22 sections, 5 theorems, 44 equations, 6 figures, 1 table.

Table of Contents

  1. Introduction
  2. Related Works
  3. Aggregation Weight Design Approach
  4. Knowledge Distillation Aided Approach
  5. Contributions
  6. System Model
  7. Image Semantic Communication
  8. Federated Learning
  9. Proposed FedSFR Algorithm
  10. Algorithm Description
  11. Quantization at $\mathcal{A}_m^{(t)}$
  12. Quantization at $\mathcal{A}_o^{(t)}$
  13. Feature Reconstruction Learning
  14. From Feature Reconstruction To Image Reconstruction
  15. First-Order Taylor Expansion
  16. ...and 7 more sections

Key Result

Lemma 1

Suppose we uniformly sample a subset $\mathcal{B}_0$ from a given set $\mathcal{B}$ without replacement. Then, the following unbiasedness property holds: $\mathbb{E}_{\mathcal{B}_0}\left[\frac{|\mathcal{B}|}{|\mathcal{B}_0|}\sum_{k\in\mathcal{B}_0}p_k x_k\right] = \sum_{k\in\mathcal{B}}p_k x_k,$ whe

Figures (6)

  • Figure 1: System model of semantic communication and FL for JSCC encoder/decoder update using the shared VQ codebook when $16$-QAM ($M = 16$).
  • Figure 2: Overall procedure of FedSFR with the numbered algorithmic steps.
  • Figure 3: PSNR of (a) the proposed scheme and the baselines, and (b) the proposed scheme with varying learning rates for CIFAR-10 dataset.
  • Figure 4: PSNR of JSCC models trained via FedSFR and FedAvg, tested over AWGN and Rayleigh fading channels for CIFAR-10 dataset.
  • Figure 5: PSNR of (a) the proposed scheme and the baselines, and (b) the proposed scheme with varying learning rates for CelebA dataset.
  • ...and 1 more figures

Theorems & Definitions (5)

  • Lemma 1: li2019convergence
  • Proposition 1: Error memory compensation
  • Theorem 1
  • Lemma 2
  • Lemma 3