This project implements a neural network from scratch in C++, inspired by Andrej Karpathy's micrograd. The implementation focuses on building an autograd engine and neural network components from first principles.

• Autograd Engine: Core implementation of automatic differentiation

  • Value class with support for basic operations (+, *, tanh)
  • Automatic gradient computation through backward propagation
  • Computation graph visualization capabilities

• Neural Network Components:

  • Neuron class implementing a single artificial neuron
  • Layer class for grouping neurons
  • Neural Network class for deep learning architectures

The Value class is the fundamental building block, implementing:

• Forward pass computation
• Backward propagation of gradients
• Topological sorting of computation graphs
• Debug visualization support

The project includes visualization tools to help understand:

• Forward pass computation steps
• Gradient flow during backpropagation
• Computation graph structure

Running the program with the visualizeValue() function demonstrates a basic Value class implementation with debug visualization enabled:

1. Clone the repository
2. Build using CMake:

   mkdir build
   cd build
   cmake -DDEBUG=ON ..   # Enable debug visualization
   # or cmake -DDEBUG=OFF .. for release build
   make

3. Run the executable:

   ./neural_network

   When built with DEBUG=ON, the program will generate visualization files for the computation graphs.