Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

COMSPLIT: A Communication-Aware Split Learning Design for Heterogeneous IoT Platforms

Vukan Ninkovic, Dejan Vukobratovic, Dragisa Miskovic, Marco Zennaro

TL;DR

Numerical results show superior performance of COMSPLIT compared to vanilla SL approaches (that assume ideal communication channel), demonstrating its ability to offer both design simplicity and adaptability to different channel conditions.

Abstract

The significance of distributed learning and inference algorithms in Internet of Things (IoT) network is growing since they flexibly distribute computation load between IoT devices and the infrastructure, enhance data privacy, and minimize latency. However, a notable challenge stems from the influence of communication channel conditions on their performance. In this work, we introduce COMSPLIT: a novel communication-aware design for split learning (SL) and inference paradigm tailored to processing time series data in IoT networks. COMSPLIT provides a versatile framework for deploying adaptable SL in IoT networks affected by diverse channel conditions. In conjunction with the integration of an early-exit strategy, and addressing IoT scenarios containing devices with heterogeneous computational capabilities, COMSPLIT represents a comprehensive design solution for communication-aware SL in IoT networks. Numerical results show superior performance of COMSPLIT compared to vanilla SL approaches (that assume ideal communication channel), demonstrating its ability to offer both design simplicity and adaptability to different channel conditions.

COMSPLIT: A Communication-Aware Split Learning Design for Heterogeneous IoT Platforms

TL;DR

Numerical results show superior performance of COMSPLIT compared to vanilla SL approaches (that assume ideal communication channel), demonstrating its ability to offer both design simplicity and adaptability to different channel conditions.

Abstract

The significance of distributed learning and inference algorithms in Internet of Things (IoT) network is growing since they flexibly distribute computation load between IoT devices and the infrastructure, enhance data privacy, and minimize latency. However, a notable challenge stems from the influence of communication channel conditions on their performance. In this work, we introduce COMSPLIT: a novel communication-aware design for split learning (SL) and inference paradigm tailored to processing time series data in IoT networks. COMSPLIT provides a versatile framework for deploying adaptable SL in IoT networks affected by diverse channel conditions. In conjunction with the integration of an early-exit strategy, and addressing IoT scenarios containing devices with heterogeneous computational capabilities, COMSPLIT represents a comprehensive design solution for communication-aware SL in IoT networks. Numerical results show superior performance of COMSPLIT compared to vanilla SL approaches (that assume ideal communication channel), demonstrating its ability to offer both design simplicity and adaptability to different channel conditions.

Paper Structure

This paper contains 29 sections, 4 equations, 12 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Background
  4. Split Learning & Inference
  5. Split Learning--Based LSTM
  6. System Model
  7. Split Inference Over Erasure Channel
  8. Split Inference Over AWGN Channel
  9. Split Inference with Early--Exit
  10. Split Inference with Heterogeneous IoT Devices
  11. COMSPLIT: Communication--Aware Split Inference Techniques
  12. Baseline COMSPLIT Model
  13. COMSPLIT Model with Early--Exit
  14. COMSPLIT with Heterogeneous IoT Devices
  15. Performance Evaluation
  16. ...and 14 more sections

Figures (12)

  • Figure 1: Split learning/inference architecture with edge device and server sub--networks implementation.
  • Figure 2: LSTM--based split learning/inference algorithms: a) LSTMSPLIT jiang_2022; b) Fedsl abedi_2023.
  • Figure 3: Communication-aware split learning/inference model incorporating: a) Erasure channel; b) AWGN channel; c) Early exit yankowski_2023.
  • Figure 4: Split learning/inference model utilizing multiple edge devices with varying computational capabilities.
  • Figure 5: Use case 2: Water quality monitoring IoT system - Initial setup with a) smart buoy and b) UAV equipped with communication equipment
  • ...and 7 more figures