Home

nuNN Library Wiki

nuNN is a compact C++20 machine learning library built for study, experimentation, and small practical demos. The project intentionally keeps the algorithms visible: forward passes, gradients, training loops, serialization, and example applications are implemented in readable C++ instead of being hidden behind a large framework.

This wiki complements the README:

• the README is the quick project overview;
• NEWS is the release history;
• the wiki is the guided documentation layer, with theory notes, diagrams, and links back to the implementation.

Quick Links

• Getting Started
• Neural Networks
• MNIST and OCR
• Recurrent Networks
• Convolutional Networks and Transformer
• Reinforcement Learning
• Examples Gallery
• Theory Notes

What nuNN Includes

nuNN includes feedforward neural networks, recurrent networks, unsupervised models, convolutional components, a small decoder-only transformer, tabular reinforcement learning, DQN, classical ML helpers, model serialization, Graphviz topology export, MNIST parsing, OCR demos, and package/install helpers.

The current implementation includes:

• Perceptron
• MlpNN
• MlpMatrixNN
• VanillaRnn
• Gru
• Lstm
• HopfieldNN
• Autoencoder
• Rbf
• Rbm
• Vae
• Som
• Conv1DLayer, MaxPool1DLayer, ConvNet
• MiniTransformer
• LinearRegression
• KMeans
• Pca
• tabular QLearn / Sarsa
• Dqn with replay buffer and target network

Using Book Material

Some diagrams in this wiki are adapted from the accompanying book material and stored as PNG assets under assets/. They are used here to connect the theory to the corresponding nuNN classes and demo programs.