Dendron: Enhancing Human Activity Recognition with On-Device TinyML Learning

2025-03-03

AI Summary

To address the challenge of continual learning for human activity recognition (HAR) on resource-constrained wearable devices under low-shot, edge-deployment conditions, this paper proposes the first embedded-edge-oriented incremental TinyML learning framework. The framework integrates a lightweight neural network architecture, meta-learning-guided parameter-efficient fine-tuning, and an embedded-friendly online inference and model update strategy. Evaluated on two public HAR datasets and deployed on the STM32-NUCLEO-F401RE microcontroller platform, it achieves >92% classification accuracy with only ≤5 labeled samples per class while maintaining a model memory footprint of <120 KB. This work establishes the first low-supervision, high-efficiency, and deployable edge-based incremental HAR learning solution, introducing a novel paradigm for practical TinyML deployment in dynamic sensing scenarios.

Technology Category

• Machine Learning: Learning on the Edge & Model Compression
• Humans and AI: Human-in-the-loop Machine Learning
• Planning, Routing, and Scheduling: Activity and Plan Recognition

Application Category

• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Human activity recognition (HAR) is a research field that employs Machine Learning (ML) techniques to identify user activities. Recent studies have prioritized the development of HAR solutions directly executed on wearable devices, enabling the on-device activity recognition. This approach is supported by the Tiny Machine Learning (TinyML) paradigm, which integrates ML within embedded devices with limited resources. However, existing approaches in the field lack in the capability for on-device learning of new HAR tasks, particularly when supervised data are scarce. To address this limitation, our paper introduces Dendron, a novel TinyML methodology designed to facilitate the on-device learning of new tasks for HAR, even in conditions of limited supervised data. Experimental results on two public-available datasets and an off-the-shelf device (STM32-NUCLEO-F401RE) show the effectiveness and efficiency of the proposed solution.

Problem

• Enables on-device learning for human activity recognition.
• Addresses limited supervised data in TinyML applications.
• Introduces Dendron for efficient HAR task adaptation.

Innovation

• On-device TinyML for activity recognition
• Learning new tasks with limited data
• Efficient implementation on embedded devices