The Hello World example

Introduction

The Hello World example demonstrates a regression model implemented as a neural network with an input layer of a single scalar value (the x parameter of y=sin(x)) , a hidden layer of 16 neurons and an output layer, again of a single scalar value (the y in y=sin(x)). To train the network a data set of 1000 values of the sin function is used onto which a small random noise is added. Once trained we can predict sin(x) for any value of x.

Like most "Hello World" type programs this is pretty useless since we can get the precise value of sin(x) very easily, but it shows the principles of

• how a neural network regression model can be implemented and trained
• how the trained model can be converted into a model that can run on a micro-controller
• and it shows that the infrastructure for building and employing ML models on the ESP32 is in place and is working correctly.

Different versions of the Hello World example found on the Internet

The example program comes in different version

• the hello_world tflite-micro version which is part of the tflite-micro github repository
• the hello_world Arduino version written in C++
• the hello_world ESP-IDF version
• the hello_world MicroPython version

-- Uli Raich - 2022-01-31

Comments