Neural-Net Implementation with Numpy

• Pure, simple deep neural-network implementation with Numpy. Supports multiple layers of any length and any number of outputs.
• Can achieve a score of around 0.9 on MNIST dataset.
• Currently uses sigmoid for activation.

TODO

• Multiple activation functions (maybe even custom if derivative is supplied).
• More implementations.
• Different types of networks: convolutional, recurrent etc..

Sources

• Make Your Own Neural Network by Tariq Rashid