Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Joint Optimization of Latency and Accuracy for Split Federated Learning in User-Centric Cell-Free MIMO Networks

Zitong Wang, Cheng Zhang, Wen Wang, Shuigen Yang, Haiming Wang, Yongming Huang

TL;DR

This work introduces UCSFL, a framework for joint optimization of training latency and accuracy in a user-centric CF-MIMO network by leveraging AP-side DPUs and a two-level model aggregation. The key innovation is defining a latency-to-accuracy ratio that ties per-iteration delay to AP-cluster size, enabling a unified objective across heterogeneous users. The optimization is tackled via a two-stage approach: a short-term NBCD algorithm for uplink power control and downlink beamforming, and a long-term MAIS DRL-based method for adaptive model splitting and AP clustering. Simulations show that UCSFL accelerates convergence of VGG16 and MNIST tasks and adaptively adjusts splitting and clustering to resource variations, achieving a lower latency-to-accuracy ratio than baselines and faster per-iteration learning dynamics. The results demonstrate the practical potential of aligning communication-computing trade-offs with distributed edge AI in next-generation networks.

Abstract

This paper proposes a user-centric split federated learning (UCSFL) framework for user-centric cell-free multiple-input multiple-output (CF-MIMO) networks to support split federated learning (SFL). In the proposed UCSFL framework, users deploy split sub-models locally, while complete models are maintained and updated at access point (AP)-side distributed processing units (DPUs), followed by a two-level aggregation procedure across DPUs and the central processing unit (CPU). Under standard machine learning (ML) assumptions, we provide a theoretical convergence analysis for UCSFL, which reveals that the AP-cluster size is a key factor influencing model training accuracy. Motivated by this result, we introduce a new performance metric, termed the latency-to-accuracy ratio, defined as the ratio of a user's per-iteration training latency to the weighted size of its AP cluster. Based on this metric, we formulate a joint optimization problem to minimize the maximum latency-to-accuracy ratio by jointly optimizing uplink power control, downlink beamforming, model splitting, and AP clustering. The resulting problem is decomposed into two sub-problems operating on different time scales, for which dedicated algorithms are developed to handle the short-term and long-term optimizations, respectively. Simulation results verify the convergence of the proposed algorithms and demonstrate that UCSFL effectively reduces the latency-to-accuracy ratio of the VGG16 model compared with baseline schemes. Moreover, the proposed framework adaptively adjusts splitting and clustering strategies in response to varying communication and computation resources. An MNIST-based handwritten digit classification example further shows that UCSFL significantly accelerates the convergence of the VGG16 model.

Joint Optimization of Latency and Accuracy for Split Federated Learning in User-Centric Cell-Free MIMO Networks

TL;DR

This work introduces UCSFL, a framework for joint optimization of training latency and accuracy in a user-centric CF-MIMO network by leveraging AP-side DPUs and a two-level model aggregation. The key innovation is defining a latency-to-accuracy ratio that ties per-iteration delay to AP-cluster size, enabling a unified objective across heterogeneous users. The optimization is tackled via a two-stage approach: a short-term NBCD algorithm for uplink power control and downlink beamforming, and a long-term MAIS DRL-based method for adaptive model splitting and AP clustering. Simulations show that UCSFL accelerates convergence of VGG16 and MNIST tasks and adaptively adjusts splitting and clustering to resource variations, achieving a lower latency-to-accuracy ratio than baselines and faster per-iteration learning dynamics. The results demonstrate the practical potential of aligning communication-computing trade-offs with distributed edge AI in next-generation networks.

Abstract

This paper proposes a user-centric split federated learning (UCSFL) framework for user-centric cell-free multiple-input multiple-output (CF-MIMO) networks to support split federated learning (SFL). In the proposed UCSFL framework, users deploy split sub-models locally, while complete models are maintained and updated at access point (AP)-side distributed processing units (DPUs), followed by a two-level aggregation procedure across DPUs and the central processing unit (CPU). Under standard machine learning (ML) assumptions, we provide a theoretical convergence analysis for UCSFL, which reveals that the AP-cluster size is a key factor influencing model training accuracy. Motivated by this result, we introduce a new performance metric, termed the latency-to-accuracy ratio, defined as the ratio of a user's per-iteration training latency to the weighted size of its AP cluster. Based on this metric, we formulate a joint optimization problem to minimize the maximum latency-to-accuracy ratio by jointly optimizing uplink power control, downlink beamforming, model splitting, and AP clustering. The resulting problem is decomposed into two sub-problems operating on different time scales, for which dedicated algorithms are developed to handle the short-term and long-term optimizations, respectively. Simulation results verify the convergence of the proposed algorithms and demonstrate that UCSFL effectively reduces the latency-to-accuracy ratio of the VGG16 model compared with baseline schemes. Moreover, the proposed framework adaptively adjusts splitting and clustering strategies in response to varying communication and computation resources. An MNIST-based handwritten digit classification example further shows that UCSFL significantly accelerates the convergence of the VGG16 model.
Paper Structure (18 sections, 1 theorem, 71 equations, 10 figures, 2 tables, 2 algorithms)

This paper contains 18 sections, 1 theorem, 71 equations, 10 figures, 2 tables, 2 algorithms.

Table of Contents

  1. Introduction
  2. System Model
  3. Wireless Transmission in User-Centric Cell-Free MIMO System
  4. User-Centric Split Federated Learning Model in Cell-Free MIMO System
  5. Convergence Analysis
  6. Joint Latency and Accuracy for UCSFL: Problem Formulation
  7. Training Latency
  8. Problem Formulation
  9. Joint Communication and Computation Resource Allocation: Proposed Algorithm
  10. Problem Splitting
  11. Solving the Short-Term Sub-problem
  12. Solving the long-Term Sub-problem
  13. Simulation and Discussion
  14. Parameters and Setup
  15. Results and Discussion
  16. ...and 3 more sections

Key Result

Theorem 1

Let $\alpha=\frac{\beta}{\mu}$, $\iota=\max\left\{8\alpha, T\right\}-1$ and choose the learning rate $\eta_t=\frac{2}{\mu\left(t+\iota\right)}$ . After $T$ iterations, there is an upper bound between the mean value and the optimal value of any device's loss function, which satisfies the following re where $R=\frac{3C_u+1}{C_u}\left(T-1\right)^2LZ^2+\frac{C_u+1}{2C_u}LZ^2+\frac{C_u+1}{2C_u}L\epsilo

Figures (10)

  • Figure 1: A user-centric CF-MIMO network to support FL.
  • Figure 2: A user-centric split federated learning (UCSFL) framework.
  • Figure 3: Convergence behavior of MAIS.
  • Figure 4: Splitting and clustering strategies versus weight coefficient. Here, we set $f^{ue}=1~\rm{GHz}$ and $w=15~\rm{kHz}$.
  • Figure 5: Latency of the worst-performing UE versus user computing frequency. Here, we set $w=15~\rm{kHz}$ and $\ell=1$.
  • ...and 5 more figures

Theorems & Definitions (2)

  • Theorem 1
  • proof