areolsen/plain-artificial-intelligence

PAI - Plain Artifical Intelligence, written in C#, from scratch.

Plain Artificial Intelligence

PAI - Plain Artificial Intelligence, written in C#, from scratch. This is a simple neural network trained on MNIST, as a learning experience, it is easily configurable in code and can be reworked to fit every need. By default, it uses, ReLU combined with Softmax as the activation functions, and Binary Cross entropy as the loss function, but other activation functions and loss functions are included. After 8 epochs it reached a percentage correctly guessed on the MNIST test data of approx. 97%. The network can be saved and loaded using Binary Serialization. In the future, this project will be attempted to be multithreaded. Feel free to add any changes to the code that may help the network learn better :)

The MNIST Dataset used for this project can be downloaded here: Kaggle

plain-artificial-intelligence's Issues

Are you still planning to add multithreading?

The attempt to add multithreading (commented out in the MBGD() function) failed because it is not thead safe.
Are you still planning to add multithreading, and if so, do you need a hand?

Please, also review my pull request - besides the cosmetics I added several small enhancements and fixes.
Let me know if you consider merging my changes. If you don't plan to, I will leave you in peace.