Energy Efficiency and Classification Locality: Pareto-optimal trade-offs in multi-class sensor-based Human Activity Recognition
Files
Date
2024-01-03
Authors
Contributor
Advisor
Department
Instructor
Depositor
Speaker
Researcher
Consultant
Interviewer
Narrator
Transcriber
Annotator
Journal Title
Journal ISSN
Volume Title
Publisher
Volume
Number/Issue
Starting Page
7730
Ending Page
Alternative Title
Abstract
Human Activity Recognition (HAR) is one of the central analytical workloads on wearable devices and effective solutions often require methods from machine learning or artificial intelligence to perform activity classification. Contrary to the hardware in wearable devices, these algorithms have mainly been developed without strict constraints on processing power, memory footprint, or storage space. Wearable applications, therefore, need to simultaneously balance at least the energy consumption of the device and the predictive performance of the algorithms. A plethora of software development kits and frameworks aim to bring those algorithms to smaller ultra-low power Systems-on-Chip (SoCs) and promise efficient execution of analytical workloads. In this study, we provide a holistic view of hardware, algorithms, and software that is useful to build smart wearable devices and provide guidance to researchers and practitioners for the selection of algorithms, configurations, and toolsets that, in combination, provide a Pareto-optimal trade-off between energy consumption and classification performance.
Description
Keywords
Software Technology and Software Development, machine learning, neural networks, pareto front, ultra low power devices
Citation
Extent
10 pages
Format
Geographic Location
Time Period
Related To
Proceedings of the 57th Hawaii International Conference on System Sciences
Related To (URI)
Table of Contents
Rights
Attribution-NonCommercial-NoDerivatives 4.0 International
Rights Holder
Local Contexts
Collections
Email libraryada-l@lists.hawaii.edu if you need this content in ADA-compliant format.