Deploying shallow Neural Networks on low power ARM Cortex M

Deploying a trained network in limited precision on an ARM microcontroller such as Arduino Uno

In this example we illustrate a MATLAB and Simulink workflow on how to train and deploy a machine learning model to a low-power microcontroller on the edge. We demonstrate how to train a shallow neural network for a regression problem, how to generate readable single precision floating point or Fixed-point code and how to deploy to an ARM cortex M microcontroller such as an Arduino Uno.
We use the engine dataset for estimating engine emission levels based on measurements of fuel consumption and speed. This is a regression problem and we use a shallow neural network to model the system.
The download contains the example dataset, the trained model exported as a MATLAB function and an equivalent Simulink model and a detailed article explaining the workflow steps. It also contains all the required scripts to automate some of the tasks.