Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Adaptive Split Learning over Energy-Constrained Wireless Edge Networks

Zuguang Li, Wen Wu, Shaohua Wu, Wei Wang

TL;DR

This work tackles the challenge of training AI models over energy-constrained wireless edge networks where device heterogeneity and fluctuating channels hamper split learning (SL). It introduces Adaptive Split Learning (ASL) and the OPEN online algorithm, which uses Lyapunov drift-plus-cost optimization to dynamically select split points and allocate edge server resources in a per-slot, information-limited manner. The method transforms a long-term stochastic mixed-integer problem into tractable per-slot subproblems, solved via a two-layer approach that combines an upper-layer resource allocation with a lower-layer split-point search. Empirical results on a 12-layer LeNet trained over MNIST show substantial reductions in training delay (up to ~54% vs SL) and energy consumption (up to ~22% vs SL) while respecting energy constraints, highlighting the practical value of adaptive split learning in edge AI deployments.

Abstract

Split learning (SL) is a promising approach for training artificial intelligence (AI) models, in which devices collaborate with a server to train an AI model in a distributed manner, based on a same fixed split point. However, due to the device heterogeneity and variation of channel conditions, this way is not optimal in training delay and energy consumption. In this paper, we design an adaptive split learning (ASL) scheme which can dynamically select split points for devices and allocate computing resource for the server in wireless edge networks. We formulate an optimization problem to minimize the average training latency subject to long-term energy consumption constraint. The difficulties in solving this problem are the lack of future information and mixed integer programming (MIP). To solve it, we propose an online algorithm leveraging the Lyapunov theory, named OPEN, which decomposes it into a new MIP problem only with the current information. Then, a two-layer optimization method is proposed to solve the MIP problem. Extensive simulation results demonstrate that the ASL scheme can reduce the average training delay and energy consumption by 53.7% and 22.1%, respectively, as compared to the existing SL schemes.

Adaptive Split Learning over Energy-Constrained Wireless Edge Networks

TL;DR

This work tackles the challenge of training AI models over energy-constrained wireless edge networks where device heterogeneity and fluctuating channels hamper split learning (SL). It introduces Adaptive Split Learning (ASL) and the OPEN online algorithm, which uses Lyapunov drift-plus-cost optimization to dynamically select split points and allocate edge server resources in a per-slot, information-limited manner. The method transforms a long-term stochastic mixed-integer problem into tractable per-slot subproblems, solved via a two-layer approach that combines an upper-layer resource allocation with a lower-layer split-point search. Empirical results on a 12-layer LeNet trained over MNIST show substantial reductions in training delay (up to ~54% vs SL) and energy consumption (up to ~22% vs SL) while respecting energy constraints, highlighting the practical value of adaptive split learning in edge AI deployments.

Abstract

Split learning (SL) is a promising approach for training artificial intelligence (AI) models, in which devices collaborate with a server to train an AI model in a distributed manner, based on a same fixed split point. However, due to the device heterogeneity and variation of channel conditions, this way is not optimal in training delay and energy consumption. In this paper, we design an adaptive split learning (ASL) scheme which can dynamically select split points for devices and allocate computing resource for the server in wireless edge networks. We formulate an optimization problem to minimize the average training latency subject to long-term energy consumption constraint. The difficulties in solving this problem are the lack of future information and mixed integer programming (MIP). To solve it, we propose an online algorithm leveraging the Lyapunov theory, named OPEN, which decomposes it into a new MIP problem only with the current information. Then, a two-layer optimization method is proposed to solve the MIP problem. Extensive simulation results demonstrate that the ASL scheme can reduce the average training delay and energy consumption by 53.7% and 22.1%, respectively, as compared to the existing SL schemes.
Paper Structure (14 sections, 2 theorems, 24 equations, 3 figures, 1 table, 1 algorithm)

This paper contains 14 sections, 2 theorems, 24 equations, 3 figures, 1 table, 1 algorithm.

Table of Contents

  1. Introduction
  2. System Model and Problem Formulation
  3. Considered Scenario
  4. Training Delay
  5. Energy Consumption
  6. Problem Formulation
  7. OPEN Algorithm
  8. Problem Transformation
  9. Upper Layer Analysis
  10. Lower layer Analysis
  11. Simulation Results
  12. Simulation Setup
  13. Performance Evaluation
  14. Conclusion

Key Result

Theorem 1

Equation eq: energy deficit queue is essentially equivalent to the long-term throughput constraint in Problem 1, if the stability condition $\lim_{T \to \infty} Q^{T}_{m,n}/T = 0, ~ \forall m\in \mathcal{M}, ~n \in \mathcal{N}$ can be satisfied.

Figures (3)

  • Figure 1: System model.
  • Figure 2: Optimal split point and computing resource allocation at each episode.
  • Figure 3: Delay and energy consumption performance with respect to different schemes.

Theorems & Definitions (2)

  • Theorem 1
  • Theorem 2