Embedded Open Source Summit 2023

As deployment of TinyML algorithms on resource-constrained embedded devices is relatively cheap in comparison to cloud based ML solutions, we experience a boom of TinyML solutions in versatile use cases including security systems, intelligent lighting, wildlife conservation, early warnings about system failure detection and many others. Despite the quickly growing number of available TinyML projects, existing solutions for MLOps are not well integrated with each other and the rest of the IoT ecosystem. As a result, IoT solution providers are required to integrate multiple communication stacks into their devices, which is challenging for resource-constrained devices. These challenges can be addressed by integration of the LwM2M communication protocol with TinyML solutions. In such a scenario, we obtain a single communication stack that multiplexes OTA updates, telemetry data collection and device management on a single network connection. Mieszko will talk about integration of Edge Impulse ML with LwM2M, how the solution can be standardized, what are the gains of the approach and showcase a demonstration of the whole TinyML operations flow based on Zephyr RTOS to detect system anomalies and failures.