Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Efficient Split Learning LSTM Models for FPGA-based Edge IoT Devices

Romina Soledad Molina, Vukan Ninkovic, Dejan Vukobratovic, Maria Liz Crespo, Marco Zennaro

TL;DR

The paper addresses efficient deployment of Split Learning–based LSTM on FPGA-edge IoT devices for time-series environmental monitoring. It combines pruning, quantization, and knowledge distillation within a split-inference framework enabled by high-level synthesis tooling (hls4ml) to tailor LSTM models for resource-constrained edges. On the Danube River dataset, the student LSTM (LSTM-DO-S) achieves substantial memory reduction (about $45.9\\times$) with only modest performance loss, while Split-A and Split-B variants demonstrate trade-offs between latency and resource usage. The results illustrate a practical, privacy-preserving approach for real-time edge ML in environmental monitoring, enabling multiple ML instances on FPGA hardware to meet diverse application requirements.

Abstract

Split Learning (SL) recently emerged as an efficient paradigm for distributed Machine Learning (ML) suitable for the Internet Of Things (IoT)-Cloud systems. However, deploying SL on resource-constrained edge IoT platforms poses a significant challenge in terms of balancing the model performance against the processing, memory, and energy resources. In this work, we present a practical study of deploying SL framework on a real-world Field-Programmable Gate Array (FPGA)-based edge IoT platform. We address the SL framework applied to a time-series processing model based on Recurrent Neural Networks (RNNs). Set in the context of river water quality monitoring and using real-world data, we train, optimize, and deploy a Long Short-Term Memory (LSTM) model on a given edge IoT FPGA platform in different SL configurations. Our results demonstrate the importance of aligning design choices with specific application requirements, whether it is maximizing speed, minimizing power, or optimizing for resource constraints.

Efficient Split Learning LSTM Models for FPGA-based Edge IoT Devices

TL;DR

The paper addresses efficient deployment of Split Learning–based LSTM on FPGA-edge IoT devices for time-series environmental monitoring. It combines pruning, quantization, and knowledge distillation within a split-inference framework enabled by high-level synthesis tooling (hls4ml) to tailor LSTM models for resource-constrained edges. On the Danube River dataset, the student LSTM (LSTM-DO-S) achieves substantial memory reduction (about ) with only modest performance loss, while Split-A and Split-B variants demonstrate trade-offs between latency and resource usage. The results illustrate a practical, privacy-preserving approach for real-time edge ML in environmental monitoring, enabling multiple ML instances on FPGA hardware to meet diverse application requirements.

Abstract

Split Learning (SL) recently emerged as an efficient paradigm for distributed Machine Learning (ML) suitable for the Internet Of Things (IoT)-Cloud systems. However, deploying SL on resource-constrained edge IoT platforms poses a significant challenge in terms of balancing the model performance against the processing, memory, and energy resources. In this work, we present a practical study of deploying SL framework on a real-world Field-Programmable Gate Array (FPGA)-based edge IoT platform. We address the SL framework applied to a time-series processing model based on Recurrent Neural Networks (RNNs). Set in the context of river water quality monitoring and using real-world data, we train, optimize, and deploy a Long Short-Term Memory (LSTM) model on a given edge IoT FPGA platform in different SL configurations. Our results demonstrate the importance of aligning design choices with specific application requirements, whether it is maximizing speed, minimizing power, or optimizing for resource constraints.

Paper Structure

This paper contains 14 sections, 1 equation, 4 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Background
  3. Split Learning
  4. Split Learning--Based LSTM
  5. RNN Deployment FPGA Acceleration: State of the Art
  6. IoT-Based Environmental Monitoring System
  7. System Model
  8. RNN Compression and FPGA Deployment for Time Series Processing
  9. System overview
  10. Performance Evaluation
  11. Dataset Generation
  12. Training, compression, and hardware generation
  13. Results
  14. Conclusion

Figures (4)

  • Figure 1: Split learning/inference pipeline with edge and server sub--networks for IoT systems.
  • Figure 2: IoT-based water quality monitoring system with a) smart buoy with FPGA and b) UAV-integrated communication equipment
  • Figure 3: Workflow for model compression and FPGA deployment.
  • Figure 4: Hardware evaluation framework.