Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Energy-Efficient Federated Learning and Migration in Digital Twin Edge Networks

Yuzhi Zhou, Yaru Fu, Zheng Shi, Howard H. Yang, Kevin Hung, Yan Zhang

TL;DR

This work tackles the privacy- and energy-aware deployment of federated learning in digital twin edge networks (DITEN) for 6G. It introduces a closed-form data utility function that predicts FL accuracy via an EM distance-based data distribution divergence and a data size term, enabling a long-term optimization of digital twin association and historical data allocation. An optimization-driven learning framework is proposed, combining a PPO-based DRL agent for edge-server association with convex optimization for data allocation, supported by FL convergence analysis and barrier-based reward design. Numerical results show that the proposed method outperforms baselines in objective value and data utility across varied network scales and data heterogeneity, highlighting the approach’s potential for energy-efficient, privacy-preserving FL in dynamic DITEN environments.

Abstract

The digital twin edge network (DITEN) is a significant paradigm in the sixth-generation wireless system (6G) that aims to organize well-developed infrastructures to meet the requirements of evolving application scenarios. However, the impact of the interaction between the long-term DITEN maintenance and detailed digital twin tasks, which often entail privacy considerations, is commonly overlooked in current research. This paper addresses this issue by introducing a problem of digital twin association and historical data allocation for a federated learning (FL) task within DITEN. To achieve this goal, we start by introducing a closed-form function to predict the training accuracy of the FL task, referring to it as the data utility. Subsequently, we carry out comprehensive convergence analyses on the proposed FL methodology. Our objective is to jointly optimize the data utility of the digital twin-empowered FL task and the energy costs incurred by the long-term DITEN maintenance, encompassing FL model training, data synchronization, and twin migration. To tackle the aforementioned challenge, we present an optimization-driven learning algorithm that effectively identifies optimized solutions for the formulated problem. Numerical results demonstrate that our proposed algorithm outperforms various baseline approaches.

Energy-Efficient Federated Learning and Migration in Digital Twin Edge Networks

TL;DR

This work tackles the privacy- and energy-aware deployment of federated learning in digital twin edge networks (DITEN) for 6G. It introduces a closed-form data utility function that predicts FL accuracy via an EM distance-based data distribution divergence and a data size term, enabling a long-term optimization of digital twin association and historical data allocation. An optimization-driven learning framework is proposed, combining a PPO-based DRL agent for edge-server association with convex optimization for data allocation, supported by FL convergence analysis and barrier-based reward design. Numerical results show that the proposed method outperforms baselines in objective value and data utility across varied network scales and data heterogeneity, highlighting the approach’s potential for energy-efficient, privacy-preserving FL in dynamic DITEN environments.

Abstract

The digital twin edge network (DITEN) is a significant paradigm in the sixth-generation wireless system (6G) that aims to organize well-developed infrastructures to meet the requirements of evolving application scenarios. However, the impact of the interaction between the long-term DITEN maintenance and detailed digital twin tasks, which often entail privacy considerations, is commonly overlooked in current research. This paper addresses this issue by introducing a problem of digital twin association and historical data allocation for a federated learning (FL) task within DITEN. To achieve this goal, we start by introducing a closed-form function to predict the training accuracy of the FL task, referring to it as the data utility. Subsequently, we carry out comprehensive convergence analyses on the proposed FL methodology. Our objective is to jointly optimize the data utility of the digital twin-empowered FL task and the energy costs incurred by the long-term DITEN maintenance, encompassing FL model training, data synchronization, and twin migration. To tackle the aforementioned challenge, we present an optimization-driven learning algorithm that effectively identifies optimized solutions for the formulated problem. Numerical results demonstrate that our proposed algorithm outperforms various baseline approaches.

Paper Structure

This paper contains 26 sections, 2 theorems, 44 equations, 10 figures, 2 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. System model
  3. System Model
  4. Federated Learning Model
  5. Data Utility Model
  6. Energy Consumption Model
  7. Migration energy cost
  8. Synchronization energy cost
  9. Computation energy cost
  10. FL Convergence Analysis and Problem Formulation
  11. FL Convergence Analysis
  12. Problem Formulation
  13. Algorithm Design
  14. MDP Reformulation
  15. State
  16. ...and 11 more sections

Key Result

Theorem 1

Given the learning rate $\chi$, global optimal weights $\omega^{t,*}$, and fine-tuned optimal weights $\omega^{t,*}_{u}$, $\forall u$, the upper bound of $F(\omega^{t,i+1}) - F(\omega^{t,*})$ and $F(\omega^{t,i+1}_{u}) - F(\omega^{t,*}_{u})$ can be given by and respectively.

Figures (10)

  • Figure 1: System model of the DITEN.
  • Figure 2: An illustration of our optimization-driven learning algorithm.
  • Figure 3: Estimation of the data utility function in (\ref{['utility']}).
  • Figure 4: Convergence performance of PPO algorithm with 20 users.
  • Figure 5: Objective value versus edge server number for different algorithms with $20$ users.
  • ...and 5 more figures

Theorems & Definitions (2)

  • Theorem 1
  • Theorem 2