Searching Neural Architectures for Sensor Nodes on IoT Gateways

TL;DR

This work tackles the challenge of enabling ML at the edge for privacy-sensitive IoT deployments by automatically designing hardware-aware neural architectures that run entirely on IoT gateways without sending data to clouds. It introduces a derivative-free, two-level HW-NAS that constructs a constrained search space $S_{\alpha}$ on the gateway and adapts to energy, time, and memory budgets, producing hardware-friendly architectures $A(k,c)$ tailored to each sensor node. The approach achieves state-of-the-art or competitive results on the Visual Wake Words and CWRU time-series datasets while operating within ultra-low-power MCU and gateway constraints, exemplified by strong performance on Raspberry Pi and STM32 targets. The methodology is open-source and emphasizes platform-aware adaptivity, enabling privacy-preserving, personalized edge ML for HIoT and IIoT applications, with demonstrated potential for broad deployment across resource-constrained IoT networks.

Abstract

This paper presents an automatic method for the design of Neural Networks (NNs) at the edge, enabling Machine Learning (ML) access even in privacy-sensitive Internet of Things (IoT) applications. The proposed method runs on IoT gateways and designs NNs for connected sensor nodes without sharing the collected data outside the local network, keeping the data in the site of collection. This approach has the potential to enable ML for Healthcare Internet of Things (HIoT) and Industrial Internet of Things (IIoT), designing hardware-friendly and custom NNs at the edge for personalized healthcare and advanced industrial services such as quality control, predictive maintenance, or fault diagnosis. By preventing data from being disclosed to cloud services, this method safeguards sensitive information, including industrial secrets and personal data. The outcomes of a thorough experimental session confirm that -- on the Visual Wake Words dataset -- the proposed approach can achieve state-of-the-art results by exploiting a search procedure that runs in less than 10 hours on the Raspberry Pi Zero 2.

Figures (3)

• Figure 1: Automatic designing of neural architectures at the edge, running HW-NAS on an IoT gateway, using the locally collected data.
• Figure 2: Example of the process leading to the search space $S_{\alpha}$
• Figure 3: A possible run of the search strategy plotted on the search plane.