MicroFlow

With the advancement of deep learning all over the world, I decided to make a small library to help bring some of it into the world of microcontrollers. Of course, tflite-micro exists which is made for this reason but it requires a lot more memory. MicroFlow makes it easy to convert pretrained Multi-layer perceptrons (MLPs) into Arduino code without a hassle with the only limitation being the amount of memory you have. The Arduino Mega can easily run a network with an architecture of {1, 16, 16, 8, 1} with tons of memory to spare! Thats more than 400 parameters! MicroFlow also supports a bunch of activation function that range from ReLU to SWISH.

I'm looking for contributors that would like to help me create new test cases and scenarios and test different Arduino boards.

A more in-depth look at the library and it's documentation can be found on my GitHub page here.

Here's a demo of the network predicting the trigonometric sin function with notable accuracy running on the Arduino Mega