Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Split Learning in 6G Edge Networks

Zheng Lin, Guanqiao Qu, Xianhao Chen, Kaibin Huang

TL;DR

The paper surveys split edge learning (SEL) as a scalable paradigm for training large models at the 6G edge, addressing the limitations of federated learning on resource-constrained devices. It articulates a 6G-oriented architectural framework, and proposes resource-efficient SL techniques including activation and weight compression and back-propagated gradient aggregation (EPSL). The work also develops single-cell and networking perspectives for resource management, proposing hierarchical and multi-hop SL, model placement, and migration strategies, while identifying open problems in convergence, asynchrony, and label privacy. Collectively, the study highlights that SEL can dramatically reduce on-device compute and communication while enabling large-scale, privacy-preserving training across distributed edge resources, with practical implications for ultra-low latency and context-aware AI at the edge.

Abstract

With the proliferation of distributed edge computing resources, the 6G mobile network will evolve into a network for connected intelligence. Along this line, the proposal to incorporate federated learning into the mobile edge has gained considerable interest in recent years. However, the deployment of federated learning faces substantial challenges as massive resource-limited IoT devices can hardly support on-device model training. This leads to the emergence of split learning (SL) which enables servers to handle the major training workload while still enhancing data privacy. In this article, we offer a brief overview of key advancements in SL and articulate its seamless integration with wireless edge networks. We begin by illustrating the tailored 6G architecture to support edge SL. Then, we examine the critical design issues for edge SL, including innovative resource-efficient learning frameworks and resource management strategies under a single edge server. Additionally, we expand the scope to multi-edge scenarios, exploring multi-edge collaboration and mobility management from a networking perspective. Finally, we discuss open problems for edge SL, including convergence analysis, asynchronous SL and U-shaped SL.

Split Learning in 6G Edge Networks

TL;DR

The paper surveys split edge learning (SEL) as a scalable paradigm for training large models at the 6G edge, addressing the limitations of federated learning on resource-constrained devices. It articulates a 6G-oriented architectural framework, and proposes resource-efficient SL techniques including activation and weight compression and back-propagated gradient aggregation (EPSL). The work also develops single-cell and networking perspectives for resource management, proposing hierarchical and multi-hop SL, model placement, and migration strategies, while identifying open problems in convergence, asynchrony, and label privacy. Collectively, the study highlights that SEL can dramatically reduce on-device compute and communication while enabling large-scale, privacy-preserving training across distributed edge resources, with practical implications for ultra-low latency and context-aware AI at the edge.

Abstract

With the proliferation of distributed edge computing resources, the 6G mobile network will evolve into a network for connected intelligence. Along this line, the proposal to incorporate federated learning into the mobile edge has gained considerable interest in recent years. However, the deployment of federated learning faces substantial challenges as massive resource-limited IoT devices can hardly support on-device model training. This leads to the emergence of split learning (SL) which enables servers to handle the major training workload while still enhancing data privacy. In this article, we offer a brief overview of key advancements in SL and articulate its seamless integration with wireless edge networks. We begin by illustrating the tailored 6G architecture to support edge SL. Then, we examine the critical design issues for edge SL, including innovative resource-efficient learning frameworks and resource management strategies under a single edge server. Additionally, we expand the scope to multi-edge scenarios, exploring multi-edge collaboration and mobility management from a networking perspective. Finally, we discuss open problems for edge SL, including convergence analysis, asynchronous SL and U-shaped SL.
Paper Structure (22 sections, 6 figures)

This paper contains 22 sections, 6 figures.

Table of Contents

  1. Introduction
  2. Background
  3. Federated learning
  4. Split learning
  5. Federated Learning v.s. Split Learning
  6. Architectural Design for Split Edge Learning
  7. Resource-efficient Split Learning Frameworks
  8. Split Learning with Activation Compression
  9. Split Learning with Weight Compression
  10. Split Learning with Back-propogated Gradient Aggregation
  11. Resource Management for Split Learning: The Single-cell Perspective
  12. Joint Resource Allocation and Model Split
  13. Client Selection
  14. Resource Management for Split Learning: The Networking Perspective
  15. Hierarchical Split Learning
  16. ...and 7 more sections

Figures (6)

  • Figure 1: An illustration of FL and the state-of-the-art SL frameworks, where model averaging in SFL is conducted on a fed server located close to clients thapa2022splitfed.
  • Figure 2: The overall architecture for SL in 6G.
  • Figure 3: The proposed EPSL scheme in lin2023efficient.
  • Figure 4: The test accuracy of ResNet-18 on HAM10000 dataset under FL, vanilla SL, SFL, PSL and EPSL. The data samples are distributed over $5$ clients under IID settings, where the total available bandwidth is 70 MHz, the computing capability of each client is uniformly distributed within $[0.1, 0.5] \times 10^9$ cycles/s, and the computing capability of the server is $7 \times 10^9$ cycles/s.
  • Figure 5: An illustration of SL under single-cell systems.
  • ...and 1 more figures