Disclaimer: This project is currently a work in progress. As such, the API has not been finalised and breaking changes are frequent.

This project can be used to create neural networks with varying architectures. It has built-in support for LSTMs, Perceptrons and Hopfield networks. You can check the tests or GoDoc to find out more.

Aims

The aim of this project is to implement a generalised neural network library in pure Go.

The goal is to allow for people to experiment and discover new ways to design neural networks. As such, the focus is on:

• Clarity: People with very basic knowledge on ANNs should be able to use this library.
• Extensibility: It should be easy to add new layers / activation functions / etc. With this in mind, this project is based off this excellent paper.

Whilst the following are considered, they are not the core goals of this project. If they contend with the core goals of this project, they will lose out:

• Speed: Parallelisation is not built-in or considered upfront.
• Memory consumption: Neurons are modelled as actual structs rather than matrix math which results in much higher memory overheads.