51. LE2ML: a microservices-based machine learning workbench as part of an agnostic, reliable and scalable architecture for smart homes
- Author
-
Florentin Thullier, Sébastien Gaboury, and Sylvain Hallé
- Subjects
Ambient intelligence ,General Computer Science ,business.industry ,Computer science ,Cloud computing ,Microservices ,Modular design ,Machine learning ,computer.software_genre ,Home automation ,Scalability ,Component-based software engineering ,Workbench ,Artificial intelligence ,business ,computer - Abstract
Over the years, several architecture of smart home has been proposed to enable the use of ambient intelligence. However, the major issue with most of them lies in their lack of high reliability and scalability. Therefore, the first contribution of this paper introduces a novel distributed architecture for smart homes, inspired by private cloud architectures, which is reliable and scalable. This implementation aims at simplifying and encouraging both the deployment of new software components as well as their reutilization to achieve the activity recognition process inside smart homes. The second contribution of this paper is the introduction of the LIARA Environment for Modular Machine Learning (LE2ML), a new machine learning workbench. Its design relies on a microservices architecture to provide a better scalability as well as smaller and faster deployments. Experiments demonstrate that our architecture is resilient to both a node failure and a total power outage. Moreover, the workbench obtained similar results, as regards the performance of the recognition, when compared to previously proposed methods.
- Published
- 2021
- Full Text
- View/download PDF