Related Experiment Video
Updated: Sep 5, 2025
03:14
Augmenting Large Language Models via Vector Embeddings to Improve Domain-Specific Responsiveness
Published on: December 6, 2024
676
Role of Data Augmentation Strategies in Knowledge Distillation for Wearable Sensor Data
Eun Som Jeon1, Anirudh Som2, Ankita Shukla1
1School of Arts, Media and Engineering and School of Electrical, Computer and Energy Engineering, Arizona State University, Tempe, AZ 85281 USA.
IEEE Internet of Things Journal
|July 11, 2022
Summary
Knowledge distillation (KD) enables smaller neural networks for wearable devices. Optimal data augmentation and network choices for KD on time-series data depend on the specific dataset, but general recommendations offer a strong baseline.
Area of Science:
- Computer Science
- Machine Learning
- Wearable Technology
Background:
- Deep neural networks (DNNs) excel at classification but their large size hinders deployment on edge devices like wearables.
- Knowledge distillation (KD) trains smaller, efficient models using larger pre-trained networks, making DNNs suitable for resource-constrained devices.
Purpose of the Study:
- This study investigates the application and challenges of KD for time-series data analysis on wearable devices.
- It explores the impact of various data augmentation strategies on KD performance for human activity recognition.
Main Methods:
- A comparative analysis of common and hybrid data augmentation techniques was performed within the KD framework.
- The study utilized diverse datasets, ranging from small public collections to a large-scale interventional study dataset.
Main Results:
- The effectiveness of data augmentation techniques in KD for time-series data varies significantly.
- Optimal network architectures and data augmentation strategies are dataset-specific for human activity analysis.
Conclusions:
- While optimal KD strategies are dataset-dependent, this research provides general recommendations for achieving robust baseline performance across different wearable datasets.
- Further research is needed to establish a coherent strategy for data augmentation selection in KD for time-series data.
