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

Force Classification01:22

Force Classification

1.2K
Forces play a crucial role in the study of physics and engineering. They are essential in describing the motion, behavior, and equilibrium of objects in the physical world. Forces can be classified based on their origin, type, and direction of action.
Contact and non-contact forces are two of the most widely used categories of forces. As the name suggests, contact forces require physical contact between two objects to act upon each other. Examples of contact forces include frictional,...
1.2K

You might also read

Related Articles

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

Sort by
Same author

Beyond the Bedside: Machine Learning-Guided Length of Stay (LOS) Prediction for Cardiac Patients in Tertiary Care.

Healthcare (Basel, Switzerland)·2024
See all related articles

Related Experiment Video

Updated: Jun 14, 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

2.6K

Achieving More with Less: A Lightweight Deep Learning Solution for Advanced Human Activity Recognition (HAR).

Sarab AlMuhaideb1, Lama AlAbdulkarim1, Deemah Mohammed AlShahrani1

  • 1Department of Computer Science, College of Computer and Information Sciences, King Saud University, P.O. Box 266, Riyadh 11362, Saudi Arabia.

Sensors (Basel, Switzerland)
|August 29, 2024
PubMed
Summary

This study replaces computationally intensive BiLSTM layers in human activity recognition models with Residual LSTMs and data augmentation. The Residual LSTM approach offers comparable accuracy with significantly reduced model complexity for efficient deployment.

Keywords:
Inertial measurement unitconvolutional neural networks (CNNs)data augmentationdeep learninghuman activity recognition (HAR)lightweight modelslong short-term memory (LSTM)sensor data

More Related Videos

A Step-by-Step Implementation of DeepBehavior, Deep Learning Toolbox for Automated Behavior Analysis
05:41

A Step-by-Step Implementation of DeepBehavior, Deep Learning Toolbox for Automated Behavior Analysis

Published on: February 6, 2020

9.4K
Deep Neural Networks for Image-Based Dietary Assessment
13:19

Deep Neural Networks for Image-Based Dietary Assessment

Published on: March 13, 2021

9.0K

Related Experiment Videos

Last Updated: Jun 14, 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

2.6K
A Step-by-Step Implementation of DeepBehavior, Deep Learning Toolbox for Automated Behavior Analysis
05:41

A Step-by-Step Implementation of DeepBehavior, Deep Learning Toolbox for Automated Behavior Analysis

Published on: February 6, 2020

9.4K
Deep Neural Networks for Image-Based Dietary Assessment
13:19

Deep Neural Networks for Image-Based Dietary Assessment

Published on: March 13, 2021

9.0K

Area of Science:

  • Computer Science
  • Artificial Intelligence
  • Machine Learning

Background:

  • Human Activity Recognition (HAR) is vital for applications in healthcare, fitness, and defense.
  • Deep learning models, especially Bidirectional Long Short-Term Memory (BiLSTM) networks, excel in HAR but are computationally demanding for resource-constrained devices.
  • The high parameter count and computational complexity of BiLSTMs limit their real-time application on devices like smartphones.

Purpose of the Study:

  • To investigate approximating the computationally intensive BiLSTM component in HAR models.
  • To explore alternative model components and data augmentation techniques to reduce computational load.
  • To maintain or improve HAR performance on resource-limited platforms.

Main Methods:

  • Replaced BiLSTM layers with Residual LSTM (ResLSTM) and standard LSTM, incorporating convolutional networks.
  • Employed data flipping augmentation to compensate for the context awareness lost by removing BiLSTM.
  • Evaluated the modified architecture on benchmark datasets: UCI-HAR, WISDM, and KU-HAR.

Main Results:

  • The ResLSTM model demonstrated superior or comparable accuracy to BiLSTM across datasets.
  • ResLSTM achieved significantly lower parameter counts and computational complexity compared to BiLSTM.
  • On UCI-HAR, ResLSTM reached 96.34% accuracy with 576,702 parameters vs. BiLSTM's 95.22% with 849,534.
  • On WISDM, ResLSTM achieved 97.20% accuracy with 192,238 parameters vs. BiLSTM's 97.23% with 283,182.
  • On KU-HAR, ResLSTM achieved 97.05% accuracy with 386,038 parameters vs. BiLSTM's 98.63% with 569,462.

Conclusions:

  • Residual LSTM networks offer an efficient alternative to BiLSTMs for HAR, maintaining high accuracy with reduced computational demands.
  • The proposed approach enables practical deployment of advanced HAR models on smartphones and other edge devices.
  • Data augmentation techniques can effectively supplement simpler architectures to replicate complex model functionalities.