Demos
Demonstrations provide researchers and practitioners with an exciting and interactive opportunity to present their systems, artifacts and/or research prototypes, either at a regular session or at the technical exhibition. In any case, it is required to avoid a commercial format, even if the demo consists of presenting a business product or service. Instead, the presentation should focus on technical aspects.
Any written support materials may be distributed locally but not published in the proceedings. Authors who already present a paper at the conference may apply for a demonstration, to complement but not to replace their paper presentation. Demonstrations can also be made by sponsor companies or as a mixed initiative involving researchers and industrial partners.
Demonstrations are based on an informal setting that encourages presenters and participants to engage in discussions about the presented work. This is an opportunity for the participants to disseminate practical results of their research and to network with other applied researchers or business partners.
Concerning the format of the demo, we can accommodate it either as a demonstration in a booth (physical area of 4 sq. meter, with a table and 2 chairs) at the exhibition area, as a poster or as a 20 min oral presentation at a session especially set up for demonstrations. It is also possible to organize the presentation of the same demo in more than one format.
Please contact the
event secretariat.
Demo on
Online Machine Learning on
Embedded Systems for the IoT
Abstract
The demo corresponds to research published in [1] and [2]. The problem is to design, for the edge, online machine learning algorithms that do not send data to the cloud, but learn on embedded systems or low-cost computers. For instance, for the smart-building sector, temperature, humidity, and CO2 sensors continually produce data, hence the requirement to continually learn about data. Dedicated strategies and algorithms are devised to exploit the lake of resources on embedded systems.
The demo is focused on online machine learning with online Kmeans clustering algorithms using data processed by ESP microcontrollers, among them an ESP-WROOM-32D microcontroller. We will use temperature, humidity, and CO2 data from a building at Grenoble University obtained every 10 minutes from 28 LoRaWAN wireless sensors.
One of the online K-Means algorithms uses an iterative technique to group unlabeled data into K clusters based on cluster centers (centroids). Each cluster's data is chosen to minimize their average distance from their respective centroid. The ARDUINO Integrated Development Environment (Arduino IDE) is the platform that we use, in one case, to develop software codes. We also showcase an experiment with Python (micropython+ulab for ESP32) as the development platform.