Minimal C++ Neural Network

This project provides a simple neural network implementation in C++, complete with:

• Matrix operations (multiply, add, transpose) using parallel STL transforms
• Activation functions (Sigmoid, ReLU, Tanh)
• Loss functions (MSE, Cross-Entropy)
• Optimizers (SGD, Momentum)
• NeuralNetwork class that ties everything together

It also includes basic test files to verify functionalities.

Features

1. Matrix (row-major) class with parallel operations using std::execution::par
2. Activation library providing Sigmoid, ReLU, Tanh
3. Loss library supporting MSE and CrossEntropy
4. Optimizers like SGD and Momentum
5. Feed-Forward Neural Network:
   • Multi-layer
   • Forward pass, backprop, momentum-based weight updates

Building

You can compile it with C++17, on Windows. A Visual Studio solution file is provided.

Why Does This Exist?

I was bored, and decided to learn how neural networks worked. This was a culmination of quite some time of knowledge.