Combining Decision Tree and Convolutional Neural Network for Energy Efficient On-Device Activity Recognition
Activity recognition and the sub-problem of motion classification can be performed by using the data provided by inertial measurements units. For many applications it is of high interest to process the data in an energy efficient manner. This can be realized by processing the data close to the sensor on a low-cost, low power embedded device. In this paper, we explore a hybrid, hierarchical architecture that combines a decision tree (DT) with a convolutional neural network (CNN). Different DTs are executed in the sensor whereas the CNN is executed on a central processing unit close to the sensor. We show to which extent the hybrid models allow to reduce the amount of required energy and how the accuracy is affected.
- Published in:
2023 IEEE 16th International Symposium on Embedded Multicore/Many-core Systems-on-Chip (MCSoC) - Type:
Inproceedings - Authors:
Brehler, Marius; Camphausen, Lucas - Year:
2023 - Source:
https://ieeexplore.ieee.org/document/10387912
Citation information
Brehler, Marius; Camphausen, Lucas: Combining Decision Tree and Convolutional Neural Network for Energy Efficient On-Device Activity Recognition, 2023 IEEE 16th International Symposium on Embedded Multicore/Many-core Systems-on-Chip (MCSoC), 2023, 179--185, December, https://ieeexplore.ieee.org/document/10387912, Brehler.Camphausen.2023a,
@Inproceedings{Brehler.Camphausen.2023a,
author={Brehler, Marius; Camphausen, Lucas},
title={Combining Decision Tree and Convolutional Neural Network for Energy Efficient On-Device Activity Recognition},
booktitle={2023 IEEE 16th International Symposium on Embedded Multicore/Many-core Systems-on-Chip (MCSoC)},
pages={179--185},
month={December},
url={https://ieeexplore.ieee.org/document/10387912},
year={2023},
abstract={Activity recognition and the sub-problem of motion classification can be performed by using the data provided by inertial measurements units. For many applications it is of high interest to process the data in an energy efficient manner. This can be realized by processing the data close to the sensor on a low-cost, low power embedded device. In this paper, we explore a hybrid, hierarchical...}}
