GEAR-FEN: Generalized Feature Representation for Kinematic Human Activity Recognition

Mehdi Ejtehadi, Jiongfan Chen, Diego Paez-Granados

Keywords: Human Activity Recognition, Transfer Learning, Machine Learning, Classification

Abstract:

This study addresses the challenge of efficient human activity recognition (HAR) with limited training data. We propose GEAR-FEN (Generalized Activity Recognition Feature Extraction Network), a novel transfer learning method that transforms kinematic motion signals into a generalized feature space. GEAR-FEN potentially outperforms the state-of-the-art in scenarios with limited training data. This was demonstrated through an evaluation across 11 public HAR datasets (encompassing number of activities ranging from 6 to 33 and number of samples per activity ranging from 8628 to 1140258), using a deep learning model based on convolutional neural networks (CNN), residual bi-directional long short-term memory (ResBiLSTM), and an attention mechanism. Furthermore, we establish the generalizability of our method through performance comparisons on an independent dataset encompassing a distinct population and diverse kinematic modalities for 8 activities, and 26121 samples per activity. These findings highlight the potential of our proposed approach in robust feature representation for HAR tasks with limited data sizes.

Primary Area: transfer learning, meta learning, and lifelong learning

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.

Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 12000