Backpropagation neural network in C++
Back propagation C++ network code for education and fun
I started this project in 1993 after reading a September 1992 article in Scientific American explaining back-propagation; “Neurons for Computers”.
https://www.scientificamerican.com/magazine/sa/1992/09-01/
There wasn’t any code available at the time, so I worked it out myself from the mathematics formulas in the article. Over the next 10
years I occasionally revisited and update the code - I learned about scaling inputs, quick prop, bias, etc. All old hat today
but not back then. This code is written in C++ and gives a nice view into the internals of backpropagation networks.
It’s single threaded and CPU based. I thought of upgrading it, but with the excellent libraries released to open source these
days, like Pytorch, Tensorflow, MXNet, it’s wouldn’t really be that satisfying.
It was very useful back in the day. Today, it’s for educational purposes. There are some small test data samples in the
Examples directory.
It can learn by example and generate predictions.
Include the network code in your project. See the NNTest and NNSolvetest sub-projects to see how to include the class code into a C++ project.
Intended for Linux but works on Windows.
This project requires the expat library
Once you fetch the repo locally, you click on mindset/nnengine/doxygen/html/index.html
and view the C++ class documentation.