Running machine-learning inference on microcontrollers.
TinyML runs quantised neural networks on microcontrollers in kilobytes of memory: keyword spotting, anomaly detection, gesture recognition. Workflow: collect data, train, quantise (int8), convert (TFLite Micro), deploy and test. Power in milliwatts.
A wake-word detector runs continuously on a coin-cell device.
TinyML brings machine learning to microcontrollers, which enables local sensing, such as detecting a sound or motion without the cloud.
Running machine-learning inference on microcontrollers.
TinyML brings machine learning to microcontrollers, which enables local sensing, such as detecting a sound or motion without the cloud.
Memory, compute and energy, so models must be small, quantised and carefully designed for the target device.
They are trained on a computer, converted to a compact format, and executed by a small inference engine on the device.
Typical applications include detecting a keyword in audio on a battery-powered device, recognising a gesture from an accelerometer on a wearable and detecting abnormal motor vibration with a small neural network.
Study "Edge computing" first (topic 88 of 104 in Electronics & Embedded). It covers the groundwork this topic builds on.
Model size vs accuracy. CMSIS-NN. Always-on sensing.
A wake-word detector runs continuously on a coin-cell device.
The next topic is "Digital signal processors", which builds directly on this one.
It is classed as Advanced in Electronics & Embedded (topic 89 of 104). It is an advanced topic. It assumes you are comfortable with the intermediate topics in Electronics & Embedded.
Want to practise? Open the interactive course for quick checks, quizzes, simulations and progress tracking.