It the recent year, generative AI models have come to dominate the discourse around artificial intelligence and machine learning. Both Large Language Models and other generative models for image/video/sound use huge deep learning models, running on expensive and energy intensive GPUs. However there are several other application areas of machine learning, operating under other contraints. One of these is the area of "TinyML", where machine learning is used to analyze sensor data on microcontroller-grade systems. A typical TinyML system is under 1 watt, under 1 MB of RAM and FLASH and under 10 USD bill-of-materials.

emlearn is an open-source project started in 2018, which provides machine learning inference implementations for microcontrollers. It is written in portable C99 code, and supports models trained with scikit-learn and Tensorflow/Keras. Since 2023 the emlearn project also provides bindings for MicroPython, a Python for microcontrollers.

In this talk we will talk about machine learning on microcontrollers; the applications, developments in the field over the last years, and current trends - both on software and hardware side. This niche of machine learning is extremely concerned with computational efficiency, and we believe that these perspectives may be useful also to developers working in different areas.