TY - GEN
T1 - Enabling machine learning across heterogeneous sensor networks with graph autoencoders
AU - Medrano, Johan
AU - Lin, Fuchun Joseph
N1 - Publisher Copyright:
© Springer Nature Switzerland AG 2019.
PY - 2019
Y1 - 2019
N2 - Machine Learning (ML) has been applied to enable many life-assisting applications, such as abnormality detection in daily routines and automatic emergency request for the solitary elderly. However, in most cases ML algorithms depend on the layout of the target Internet of Things (IoT) sensor network. Hence, to deploy an application across Heterogeneous Sensor Networks (HSNs), i.e. sensor networks with different sensors type or layouts, it is required to repeat the process of data collection and ML algorithm training. In this paper, we introduce a novel framework leveraging deep learning for graphs to enable using the same activity recognition system across HSNs deployed in different smart homes. Using our framework, we were able to transfer activity classifiers trained with activity labels on a source HSN to a target HSN, reaching about 75% of the baseline accuracy on the target HSN without using target activity labels. Moreover, our model can quickly adapt to unseen sensor layouts, which makes it highly suitable for the gradual deployment of real-world ML-based applications. In addition, we show that our framework is resilient to suboptimal graph representations of HSNs.
AB - Machine Learning (ML) has been applied to enable many life-assisting applications, such as abnormality detection in daily routines and automatic emergency request for the solitary elderly. However, in most cases ML algorithms depend on the layout of the target Internet of Things (IoT) sensor network. Hence, to deploy an application across Heterogeneous Sensor Networks (HSNs), i.e. sensor networks with different sensors type or layouts, it is required to repeat the process of data collection and ML algorithm training. In this paper, we introduce a novel framework leveraging deep learning for graphs to enable using the same activity recognition system across HSNs deployed in different smart homes. Using our framework, we were able to transfer activity classifiers trained with activity labels on a source HSN to a target HSN, reaching about 75% of the baseline accuracy on the target HSN without using target activity labels. Moreover, our model can quickly adapt to unseen sensor layouts, which makes it highly suitable for the gradual deployment of real-world ML-based applications. In addition, we show that our framework is resilient to suboptimal graph representations of HSNs.
KW - Graph autoencoders
KW - Heterogeneous sensor networks
KW - Smart homes
UR - http://www.scopus.com/inward/record.url?scp=85076304107&partnerID=8YFLogxK
U2 - 10.1007/978-3-030-34255-5_11
DO - 10.1007/978-3-030-34255-5_11
M3 - Conference contribution
AN - SCOPUS:85076304107
SN - 9783030342548
T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
SP - 153
EP - 169
BT - Ambient Intelligence - 15th European Conference, AmI 2019, Proceedings
A2 - Chatzigiannakis, Ioannis
A2 - De Ruyter, Boris
A2 - Mavrommati, Irene
PB - Springer
T2 - 15th European Conference on Ambient Intelligence, AmI 2019
Y2 - 13 November 2019 through 15 November 2019
ER -