Here, you'll find a fully connected neural network, convolution, and pooling manually implemented and separated into multiple classes to make testing of each model easier.
Our primary objective is to identify the most efficient model for classifying MNIST digits and running it on resource-constrained platforms like Arduino. To achieve this, we have also utilized compression techniques to make it possible to run the model on such platforms.
In this repository you can find the model that was implemnted on Arduino using C++. The model was able to classify MNIST digits with 85% accuracy.