Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

MAGNETO: Edge AI for Human Activity Recognition -- Privacy and Personalization

Jingwei Zuo, George Arvanitakis, Mthandazo Ndhlovu, Hakim Hacid

TL;DR

MAGNETO proposes a privacy-preserving Edge AI platform for Human Activity Recognition that shifts HAR tasks from the Cloud to on-device processing. It utilizes a two-stage workflow: offline Cloud Initialization to build a class-separable embedding and a compact $NCM$ classifier, and online Edge Inference with incremental learning to accommodate new activities using a few-shot approach aided by a contrastive loss $L_c$ and distillation loss $L_d$. The system demonstrates real-time, on-device inference and on-the-fly personalization without transferring user data to the Cloud, validated via an Android demonstration with a large edge-derived dataset and a compact, low-footprint deployment (~5 MB). This approach delivers low latency, strong privacy, and high personalization, with potential applicability beyond HAR to other time-series and multimodal tasks. The work highlights a practical pathway for on-device continual learning and personalized sensing in health, fitness, and assistant applications.

Abstract

Human activity recognition (HAR) is a well-established field, significantly advanced by modern machine learning (ML) techniques. While companies have successfully integrated HAR into consumer products, they typically rely on a predefined activity set, which limits personalizations at the user level (edge devices). Despite advancements in Incremental Learning for updating models with new data, this often occurs on the Cloud, necessitating regular data transfers between cloud and edge devices, thus leading to data privacy issues. In this paper, we propose MAGNETO, an Edge AI platform that pushes HAR tasks from the Cloud to the Edge. MAGNETO allows incremental human activity learning directly on the Edge devices, without any data exchange with the Cloud. This enables strong privacy guarantees, low processing latency, and a high degree of personalization for users. In particular, we demonstrate MAGNETO in an Android device, validating the whole pipeline from data collection to result visualization.

MAGNETO: Edge AI for Human Activity Recognition -- Privacy and Personalization

TL;DR

MAGNETO proposes a privacy-preserving Edge AI platform for Human Activity Recognition that shifts HAR tasks from the Cloud to on-device processing. It utilizes a two-stage workflow: offline Cloud Initialization to build a class-separable embedding and a compact classifier, and online Edge Inference with incremental learning to accommodate new activities using a few-shot approach aided by a contrastive loss and distillation loss . The system demonstrates real-time, on-device inference and on-the-fly personalization without transferring user data to the Cloud, validated via an Android demonstration with a large edge-derived dataset and a compact, low-footprint deployment (~5 MB). This approach delivers low latency, strong privacy, and high personalization, with potential applicability beyond HAR to other time-series and multimodal tasks. The work highlights a practical pathway for on-device continual learning and personalized sensing in health, fitness, and assistant applications.

Abstract

Human activity recognition (HAR) is a well-established field, significantly advanced by modern machine learning (ML) techniques. While companies have successfully integrated HAR into consumer products, they typically rely on a predefined activity set, which limits personalizations at the user level (edge devices). Despite advancements in Incremental Learning for updating models with new data, this often occurs on the Cloud, necessitating regular data transfers between cloud and edge devices, thus leading to data privacy issues. In this paper, we propose MAGNETO, an Edge AI platform that pushes HAR tasks from the Cloud to the Edge. MAGNETO allows incremental human activity learning directly on the Edge devices, without any data exchange with the Cloud. This enables strong privacy guarantees, low processing latency, and a high degree of personalization for users. In particular, we demonstrate MAGNETO in an Android device, validating the whole pipeline from data collection to result visualization.
Paper Structure (16 sections, 8 figures)

This paper contains 16 sections, 8 figures.

Table of Contents

  1. Introduction
  2. Related work
  3. Edge Machine Learning (Edge ML)
  4. Human Activity Recognition (HAR)
  5. Proposal: MAGNETO
  6. Architecture
  7. Cloud Initialization
  8. Edge Inference and new activities learning
  9. About the Demonstration
  10. Demonstration settings
  11. Demonstration environment
  12. Dataset description
  13. Demonstration scenarios
  14. Real-time Inference
  15. Incremental Learning
  16. ...and 1 more sections

Figures (8)

  • Figure 1: left) Human Activity Recognition (HAR) protocols: left) Cloud-based Approach, constant communication between Cloud and Edge, right) Edge-based Approach, only data transfer from Cloud to Edge is allowed, showing stronger privacy guarantees.
  • Figure 2: Global system architecture of MAGNETO
  • Figure : (a)
  • Figure : (a)
  • Figure : (b)
  • ...and 3 more figures

Theorems & Definitions (2)

  • Definition 1
  • Definition 2