Neural Network Implementation in Zig

A simple neural network implementation in Zig

Features

• Multi-layer perceptron with configurable architecture
• Sigmoid activation function
• Backpropagation training algorithm
• Custom tensor operations
• Loss function implementation

Project Structure

src/
├── main.zig         # Example usage and XOR problem solver
├── network.zig      # Neural network implementation
├── tensor.zig       # Tensor operations
├── activations.zig  # Activation functions
└── loss.zig         # Loss functions

Building and Running

# Build the project
zig build

# Run the neural network
./zig-out/bin/neuralnet

Example Usage

The main program demonstrates training a neural network to solve the XOR problem:

• Architecture: 2 inputs → 4 hidden neurons → 1 output
• Training data: XOR truth table
• Learning rate: 0.5
• Epochs: 5000

The network learns to map:

• [0, 0] → 0
• [0, 1] → 1
• [1, 0] → 1
• [1, 1] → 0

Requirements

• Zig compiler (latest stable version)