Machine learning (ML) algorithms are moving processing to the IoT device due to challenges with latency, power consumption, cost, network, bandwidth, reliability, security, and more. 

As a result, interest is growing in developing neural network (NN) solutions to deploy ML on low-power IoT devices, for example with microcontrollers powered by proven Arm Cortex-M technology.

To help developers get a head start, Arm offers CMSIS-NN, an open-source library of optimized software kernels that maximize NN performance on Cortex-M processors with minimal memory overhead.

This guide to ML on Cortex-M microcontrollers offers methods for NN architecture exploration using image classification on a sample CIFAR-10 dataset to develop models that fit on power and cost-constrained IoT devices.

What's included in this guide?

• Techniques to perform NN model search within a set of typical compute constraints of microcontroller devices
• Methods to use to optimize the NN kernels in CMSIS-NN
• Ways to maximize NN performance on Cortex-M processors with the lowest memory footprint