A simple neural network implementation from scratch in C++ (STL is used only).
Clear OOP design, training algorithms were represented in the (well readable) matrix form.
Methods:
- base backpropagarion method was implemented.
- multitheading resilient back propagation (Rprop) was implemented.
Train data sets file's formats:
- simple handwriting digits (32x32) in the binary form (~950 samples).
- mnist handwriting digits (24x24) - 60000 samples.
TODO
- convolutional neural network