Semi or Unsupervised Human Activity Recognition Using Deep Learning
Principal Investigator
Name
Jiwon Lee
Institution
University of Cincinnati
Position Title
Ph.d. Candidate
Email
About this CDAS Project
Study
IDATA
(Learn more about this study)
Project ID
IDATA-85
Initial CDAS Request Approval
Oct 18, 2024
Title
Semi or Unsupervised Human Activity Recognition Using Deep Learning
Summary
We are focusing on the unsupervised recognition of human activity using data from Actigraph devices. In our initial research, we successfully identified active and resting states without the use of activity labels. However, this approach was limited by the inability to differentiate specific activities within those states. We now aim to extend this research to recognize distinct activities. To achieve this, we plan to incorporate reference activity labels from a separate dataset collected with the same Actigraph device, allowing us to both leverage that information and evaluate the performance of our semi-supervised or unsupervised methods.
Aims
- Leverage features extracted from labeled data to assign activity labels to the current unlabeled dataset.
- Assess model performance using the available labeled data as a benchmark.
- Implement deep learning or machine learning models to improve computational efficiency.
- Develop and publish a user-friendly package online for practitioners to apply our methods.
Collaborators
Jiwon Lee (University of Cincinnati)
Hang Joon Kim (University of Cincinnati)