Efficient Human Activity Recognition on Wearable Devices Using Knowledge Distillation Techniques

• DOI: 10.3390/electronics13183612
• Publication Date: 2024-09-11T14:33:01Z
• Authors: Paulo H. N. Gonça...; Hendrio Bragança; Eduardo Souto
• Keywords: Activity Recognition

Mobile and wearable devices have revolutionized the field of continuous user activity monitoring. However, analyzing vast intricate data captured by sensors these poses significant challenges. Deep neural networks shown remarkable accuracy in Human Activity Recognition (HAR), but their application on mobile is constrained limited computational resources. To address this limitation, we propose a novel method called Knowledge Distillation for (KD-HAR) that leverages knowledge distillation technique to compress deep network models HAR using inertial sensor data. Our approach transfers acquired from high-complexity teacher (state-of-the-art models) student with reduced complexity. This compression strategy allows us maintain performance while keeping costs low. assess capabilities our approach, evaluate it two popular databases (UCI-HAR WISDM) comprising smartphones. results demonstrate achieves competitive accuracy, even at rates ranging 18 42 times number parameters compared original model.