Welcome to the Realtime Artificial Neural Tool (RANT). RANT is a lightweight implementation of a Deep Learning library for C++. It also includes a Python interface. The principles underpinning its implementation are described in the book, “Demystifying Deep Learning,” (Wiley and IEEE Press, ISBN 1394205600).

RANT includes all the tools required to train and deploy Deep Learning models. It is suitable for high-performance in resource poor environments such as the edge, servers optimized for a specific task (e.g. a storage server). These systems often do not even have Python installed. RANT offers low memory consumption during both training and inference. Inference is orders of magnitude faster than most systems as the call stack is shallow and optimized. It was designed for CPUs.

RANT supports regression and classification (both multiclass and multilabel). There are many examples of Deep Learning and how to train models using RANT in the Examples folder using both the Python and C++ interface. For example, a tiny model that can learn MNIST.

Bug reports and suggestions welcome.