TinyML is the practice of fitting trained models onto a microcontroller, the same class of chip that reads sensors and drives a radio in a battery-powered device. Models are shrunk with quantization and pruning until they fit in a few hundred kilobytes of flash and run in milliseconds. Because the chip is already present, adding intelligence costs almost nothing in hardware and very little in energy.
Typical workloads are classification of short signals: a wake word in audio, a gesture from an accelerometer, or an unusual vibration signature. The limitation is capacity; a TinyML model cannot run a large vision network, and training still happens on a workstation. The guide TinyML: machine learning on microcontrollers walks through the toolchain.
Part of the IoT Glossary. Spotted an error or a missing term? Tell us.