Artificial Neural Networks from scratch

In this series, I will explain how artificial neural networks (ANN) work by going through the math.

A neural network is a supervised machine learning technique that can be used for classification or regression problems. I will focus on using them for classification problems.

There are many good libraries to implement ANN’s architecture and calculate the gradients automatically. The most common libraries are Keras + TensorFlow and Pytorch.

I propose to implement simple ANN from scratch only using the Numpy library. The motivation is to have a better understanding of how ANN work.

Here is a nice visualisation of what’s happening in an ANN during training.