Jove
Visualize
Contact Us
JoVE
x logofacebook logolinkedin logoyoutube logo
ABOUT JoVE
OverviewLeadershipBlogJoVE Help Center
AUTHORS
Publishing ProcessEditorial BoardScope & PoliciesPeer ReviewFAQSubmit
LIBRARIANS
TestimonialsSubscriptionsAccessResourcesLibrary Advisory BoardFAQ
RESEARCH
JoVE JournalMethods CollectionsJoVE Encyclopedia of ExperimentsArchive
EDUCATION
JoVE CoreJoVE BusinessJoVE Science EducationJoVE Lab ManualFaculty Resource CenterFaculty Site
Terms & Conditions of Use
Privacy Policy
Policies

Related Concept Videos

You might also read

Related Articles

Articles linked to this work by shared authors, journal, and citation graph.

Sort by
Same author

Condensate droplet roaming on nanostructured superhydrophobic surfaces.

Nature communications·2025
Same author

Exploring machine learning for untargeted metabolomics using molecular fingerprints.

Computer methods and programs in biomedicine·2024
Same author

Metal Surface Engineering for Extreme Sustenance of Jumping Droplet Condensation.

Langmuir : the ACS journal of surfaces and colloids·2023
Same author

Durable, Ultrathin, and Antifouling Polymer Brush Coating for Efficient Condensation Heat Transfer.

ACS applied materials & interfaces·2023
Same author

Nrf2-Mediated Pathway Activated by <i>Prunus spinosa</i> L. (Rosaceae) Fruit Extract: Bioinformatics Analyses and Experimental Validation.

Nutrients·2023
Same author

Out-of-Plane Biphilic Surface Structuring for Enhanced Capillary-Driven Dropwise Condensation.

Langmuir : the ACS journal of surfaces and colloids·2023

Related Experiment Video

Updated: Sep 27, 2025

Author Spotlight: Addressing Technical and Subjective Challenges in Measuring Classroom Attention
06:37

Author Spotlight: Addressing Technical and Subjective Challenges in Measuring Classroom Attention

Published on: December 15, 2023

4.2K

Exploring Artificial Neural Networks Efficiency in Tiny Wearable Devices for Human Activity Recognition.

Emanuele Lattanzi1, Matteo Donati1, Valerio Freschi1

  • 1Department of Pure and Applied Sciences, University of Urbino Piazza della Repubblica 13, 61029 Urbino, Italy.

Sensors (Basel, Switzerland)
|April 12, 2022
PubMed
Summary

Tiny wearable devices and machine learning offer opportunities for pervasive computing. A multilayer Perceptron network on a low-power device significantly reduces memory and energy consumption for human activity recognition.

Keywords:
artificial neural networkshuman activity recognitionmachine learningwearable devices

More Related Videos

Evaluation of a Smartphone-based Human Activity Recognition System in a Daily Living Environment
06:49

Evaluation of a Smartphone-based Human Activity Recognition System in a Daily Living Environment

Published on: December 11, 2015

9.0K
A Real-Time Wearable Electromyography Measurement System for Small Animals
05:00

A Real-Time Wearable Electromyography Measurement System for Small Animals

Published on: November 15, 2024

880

Related Experiment Videos

Last Updated: Sep 27, 2025

Author Spotlight: Addressing Technical and Subjective Challenges in Measuring Classroom Attention
06:37

Author Spotlight: Addressing Technical and Subjective Challenges in Measuring Classroom Attention

Published on: December 15, 2023

4.2K
Evaluation of a Smartphone-based Human Activity Recognition System in a Daily Living Environment
06:49

Evaluation of a Smartphone-based Human Activity Recognition System in a Daily Living Environment

Published on: December 11, 2015

9.0K
A Real-Time Wearable Electromyography Measurement System for Small Animals
05:00

A Real-Time Wearable Electromyography Measurement System for Small Animals

Published on: November 15, 2024

880

Area of Science:

  • Computer Science
  • Electrical Engineering
  • Biomedical Engineering

Background:

  • The proliferation of wearable devices and advancements in machine learning (ML) enable sophisticated edge computing.
  • Deploying ML inference on edge devices offers benefits like improved responsiveness, reduced energy use, and enhanced privacy.
  • Constraints of small, low-power wearable devices present significant design challenges for computational, memory, and energy demands.

Purpose of the Study:

  • To empirically investigate the trade-offs between computational resources and performance for ML models on wearable devices.
  • To characterize memory usage, energy consumption, and execution time of different neural network architectures for human activity recognition.
  • To evaluate the feasibility of running complex ML tasks on resource-constrained edge platforms.

Main Methods:

  • Utilized a public human activity recognition dataset for training and evaluation.
  • Implemented and analyzed both multilayer Perceptron (MLP) and convolutional neural network (CNN) models.
  • Characterized memory footprint, energy consumption, and execution time on a typical low-power wearable device.
  • Derived Pareto curves to visualize the trade-offs between model complexity, resource usage, and accuracy.

Main Results:

  • Demonstrated significant reductions in resource requirements for ML models on wearable devices.
  • Achieved a 4× reduction in memory usage and a 36× reduction in energy consumption with an MLP network compared to CNN models.
  • These reductions were achieved while maintaining comparable accuracy levels.
  • Pareto curves illustrate the efficiency gains achievable with simpler network architectures.

Conclusions:

  • Simpler neural network architectures, such as MLPs, offer substantial advantages in memory and energy efficiency for human activity recognition on wearable devices.
  • Edge deployment of ML for wearables is feasible by carefully selecting appropriate model architectures to balance performance and resource constraints.
  • The findings provide valuable insights for designing efficient pervasive computing applications leveraging wearable technology.