Details
| Original language | English |
|---|---|
| Title of host publication | KI 2018 |
| Subtitle of host publication | Advances in Artificial Intelligence - 41st German Conference on AI, 2018, Proceedings |
| Editors | Anni-Yasmin Turhan, Frank Trollmann |
| Publisher | Springer Verlag |
| Pages | 243-257 |
| Number of pages | 15 |
| ISBN (print) | 9783030001100 |
| Publication status | Published - 2018 |
| Event | 41st German Conference on Artificial Intelligence, KI 2018 - Berlin, Germany Duration: 24 Sept 2018 → 28 Sept 2018 |
Publication series
| Name | Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) |
|---|---|
| Volume | 11117 LNAI |
| ISSN (Print) | 0302-9743 |
| ISSN (electronic) | 1611-3349 |
Abstract
Many Deep Neural Networks (DNNs) are implemented with the single objective to achieve high classification scores. However, there can be additional objectives like the minimization of computational costs. This is especially important in the field of mobile computing where not only the computational power itself is a limiting factor but also each computation consumes energy affecting the battery life. Unfortunately, the determination of minimal structures is not straightforward. In our paper, we present a new approach to determine DNNs employing reduced structures. The networks are determined by an Evolutionary Algorithm (EA). After the DNN is trained, the EA starts to remove neurons from the network. Thereby, the fitness function of the EA is depending on the accuracy of the DNN. Thus, the EA is able to control the influence of each individual neuron. We introduce our new approach in detail. Thereby, we employ motion data recorded by accelerometer and gyroscope sensors of a mobile device. The data are recorded while drawing Japanese characters in the air in a learning context. The experimental results show that our approach is capable to determine reduced networks with similar performance to the original ones. Additionally, we show that the reduction can improve the accuracy of a network. We analyze the reduction in detail. Further, we present arising structures of the reduced networks.
Keywords
- Deep learning, Evolutionary Algorithm, Japanese characters, Motion sensor data, Neuroevolution, Pruning
ASJC Scopus subject areas
- Mathematics(all)
- Theoretical Computer Science
- Computer Science(all)
- General Computer Science
Cite this
- Standard
- Harvard
- Apa
- Vancouver
- BibTeX
- RIS
KI 2018: Advances in Artificial Intelligence - 41st German Conference on AI, 2018, Proceedings. ed. / Anni-Yasmin Turhan; Frank Trollmann. Springer Verlag, 2018. p. 243-257 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 11117 LNAI).
Research output: Chapter in book/report/conference proceeding › Conference contribution › Research › peer review
}
TY - GEN
T1 - Evolutionary structure minimization of deep neural networks for motion sensor data
AU - Lückehe, Daniel
AU - Veith, Sonja
AU - von Voigt, Gabriele
N1 - Publisher Copyright: © Springer Nature Switzerland AG 2018.
PY - 2018
Y1 - 2018
N2 - Many Deep Neural Networks (DNNs) are implemented with the single objective to achieve high classification scores. However, there can be additional objectives like the minimization of computational costs. This is especially important in the field of mobile computing where not only the computational power itself is a limiting factor but also each computation consumes energy affecting the battery life. Unfortunately, the determination of minimal structures is not straightforward. In our paper, we present a new approach to determine DNNs employing reduced structures. The networks are determined by an Evolutionary Algorithm (EA). After the DNN is trained, the EA starts to remove neurons from the network. Thereby, the fitness function of the EA is depending on the accuracy of the DNN. Thus, the EA is able to control the influence of each individual neuron. We introduce our new approach in detail. Thereby, we employ motion data recorded by accelerometer and gyroscope sensors of a mobile device. The data are recorded while drawing Japanese characters in the air in a learning context. The experimental results show that our approach is capable to determine reduced networks with similar performance to the original ones. Additionally, we show that the reduction can improve the accuracy of a network. We analyze the reduction in detail. Further, we present arising structures of the reduced networks.
AB - Many Deep Neural Networks (DNNs) are implemented with the single objective to achieve high classification scores. However, there can be additional objectives like the minimization of computational costs. This is especially important in the field of mobile computing where not only the computational power itself is a limiting factor but also each computation consumes energy affecting the battery life. Unfortunately, the determination of minimal structures is not straightforward. In our paper, we present a new approach to determine DNNs employing reduced structures. The networks are determined by an Evolutionary Algorithm (EA). After the DNN is trained, the EA starts to remove neurons from the network. Thereby, the fitness function of the EA is depending on the accuracy of the DNN. Thus, the EA is able to control the influence of each individual neuron. We introduce our new approach in detail. Thereby, we employ motion data recorded by accelerometer and gyroscope sensors of a mobile device. The data are recorded while drawing Japanese characters in the air in a learning context. The experimental results show that our approach is capable to determine reduced networks with similar performance to the original ones. Additionally, we show that the reduction can improve the accuracy of a network. We analyze the reduction in detail. Further, we present arising structures of the reduced networks.
KW - Deep learning
KW - Evolutionary Algorithm
KW - Japanese characters
KW - Motion sensor data
KW - Neuroevolution
KW - Pruning
UR - http://www.scopus.com/inward/record.url?scp=85054523618&partnerID=8YFLogxK
U2 - 10.1007/978-3-030-00111-7_21
DO - 10.1007/978-3-030-00111-7_21
M3 - Conference contribution
AN - SCOPUS:85054523618
SN - 9783030001100
T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
SP - 243
EP - 257
BT - KI 2018
A2 - Turhan, Anni-Yasmin
A2 - Trollmann, Frank
PB - Springer Verlag
T2 - 41st German Conference on Artificial Intelligence, KI 2018
Y2 - 24 September 2018 through 28 September 2018
ER -