Decision support systems require accurate decision support and intuitive interfaces to be accepted by users. When applied to IoT devices residing in varied environments and exists in multitudes of configurations, the decision support needs to be derived at an individual basis and the deployment must be done cross-platform. In this talk, we will explain how individual device-based decision support can be created using Ekkono’s embedded machine learning library and how it can be bundled with an intuitive QT interface and deployed cross-platform using QT’s framework. Ekkono’s machine learning library has a very small footprint and can even do advanced machine learning on devices such as ARM’s M0+. Furthermore, it is not only possible to do inference on pre-trained models, but also to learn on the device. Online learning is a crucial element for providing individual-based decision support since it enables the model to adapt to the device and learn its normal behavior. By creating one model per device, the machine learning task is simplified since the model only needs to explain what is normal for a particular device and not why it differs from others. In this way, Ekkono can for example create health indicators that will tell the operator when a device needs attention. However, to be accepted by users, decision support systems need intuitive interfaces and integrating Ekkono’s model within a QT application can be easily achieved. At the same time, QT’s cross-platform capabilities enable deployment to a heterogeneous fleet of devices which is often a reality in the IoT space. In this talk, we will present Ekkono’s on-device learning capabilities and how Ekkono has been integrated into QT’s framework to enable fast development of accurate and intuitive decision support systems.
Speaker: Rikard König (Co-founder and CTO, Ekkono Solutions)