This is a simple neural network that can classify handwritten digits.
The network only uses python's included libreries (mainly numpy), and takes MNIST database for training and benchmark.
This is a simple project I started on summer vacation as I wanted to learn about neural networks.
It is based on the first two chapters of Neural Netrorks and Deep Learning by Michael Nielsen, which is awesome and also free to read.
Was also inspired by the 3Blue1Brown video series Neural Networks by Grant Sanderson.
- Load MNIST images and labels
- Create a neural network and train it no MNIST data (or other data)
- Save and load networks to json file
- Create and train a new network from console UI
- Load and prepare an image to be detected by the network (in progress)
- Simple console UI (in progress)
- Improve the network's training based on the next chapters in the book
- Auto-crop loaded images to match MNIST format