Neural Networks from Scratch

Implementing an Artificial Neural Network from scratch using C++

There are two implementations listed here:

1. A basic_ann without using any libraries, just raw C++. The main function here is to train the network to act as a 3-input XOR operator.

2. A vectorized implementation of an ANN (include this header) using the ArrayFire library that allows the use of an unified source-code to compile programs for CPU, CUDA, as well as OpenCL. Some features that allow for smooth experimentation:

   • Ability to add custom activation functions using the Layer::setNewActivation function, as shown here.
   • Decaying learning rate across epochs as shown here, by defining a custom calc_LRdecay function.

Implementations

• Implementation of XOR gate with loss-values.
• Using the ANN to classify NLTK Movie-Reviews as postive or negative, by using TFIDF transform for feature description. repo