Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Federated Contrastive Learning for Personalized Semantic Communication

Yining Wang, Wanli Ni, Wenqiang Yi, Xiaodong Xu, Ping Zhang, Arumugam Nallanathan

TL;DR

Simulation results verify the superiority of the proposed FedCL framework compared to other distributed learning benchmarks in terms of task performance and robustness under different numbers of clients and channel conditions, especially in low signal-to-noise ratio and highly heterogeneous data scenarios.

Abstract

In this letter, we design a federated contrastive learning (FedCL) framework aimed at supporting personalized semantic communication. Our FedCL enables collaborative training of local semantic encoders across multiple clients and a global semantic decoder owned by the base station. This framework supports heterogeneous semantic encoders since it does not require client-side model aggregation. Furthermore, to tackle the semantic imbalance issue arising from heterogeneous datasets across distributed clients, we employ contrastive learning to train a semantic centroid generator (SCG). This generator obtains representative global semantic centroids that exhibit intra-semantic compactness and inter-semantic separability. Consequently, it provides superior supervision for learning discriminative local semantic features. Additionally, we conduct theoretical analysis to quantify the convergence performance of FedCL. Simulation results verify the superiority of the proposed FedCL framework compared to other distributed learning benchmarks in terms of task performance and robustness under different numbers of clients and channel conditions, especially in low signal-to-noise ratio and highly heterogeneous data scenarios.

Federated Contrastive Learning for Personalized Semantic Communication

TL;DR

Simulation results verify the superiority of the proposed FedCL framework compared to other distributed learning benchmarks in terms of task performance and robustness under different numbers of clients and channel conditions, especially in low signal-to-noise ratio and highly heterogeneous data scenarios.

Abstract

In this letter, we design a federated contrastive learning (FedCL) framework aimed at supporting personalized semantic communication. Our FedCL enables collaborative training of local semantic encoders across multiple clients and a global semantic decoder owned by the base station. This framework supports heterogeneous semantic encoders since it does not require client-side model aggregation. Furthermore, to tackle the semantic imbalance issue arising from heterogeneous datasets across distributed clients, we employ contrastive learning to train a semantic centroid generator (SCG). This generator obtains representative global semantic centroids that exhibit intra-semantic compactness and inter-semantic separability. Consequently, it provides superior supervision for learning discriminative local semantic features. Additionally, we conduct theoretical analysis to quantify the convergence performance of FedCL. Simulation results verify the superiority of the proposed FedCL framework compared to other distributed learning benchmarks in terms of task performance and robustness under different numbers of clients and channel conditions, especially in low signal-to-noise ratio and highly heterogeneous data scenarios.
Paper Structure (7 sections, 1 theorem, 18 equations, 4 figures, 1 algorithm)

This paper contains 7 sections, 1 theorem, 18 equations, 4 figures, 1 algorithm.

Table of Contents

  1. Introduction
  2. System Model
  3. FedCL Framework
  4. Contrastive Learning-Based SCG
  5. Convergence Analysis
  6. Simulation Results
  7. Conclusion

Key Result

Theorem 1

(One-round deviation bound). Let Assumptions a1 to a4 hold. For arbitrary client after each round, it satisfies,

Figures (4)

  • Figure 1: (a) Architecture of the proposed FedCL for multi-user semantic learning; (b) Workflow of the proposed FedCL framework.
  • Figure 2: (a) Learning performance of different schemes; (b)Task performance under different SNR with 20 clients.
  • Figure 3: (a) Learning performance under varying data semantic heterogeneity $m$ with 5 clients; (b) Task performance of different schemes with varying $m$.
  • Figure 4: The t-SNE visualization of semantic representations obtained by (a) proposed FedCL framework; (b) FedProto; (c) FedAvg.

Theorems & Definitions (1)

  • Theorem 1