Neural Network Function Learning

A visualization of a neural network learning to fit mathematical functions using Three.js.

Overview

This application uses a feedforward neural network to approximate mathematical functions, with real-time visualization using Three.js to show the learning process.

Live Demo

Visit loehx.github.io/curve-fitting-neural-network to see it in action.

Features

• Interactive 3D visualization
• Real-time training visualization
• Configurable network architecture
• Supports arbitrary functions
• Smooth animation

Technical Details

• Multiple configurable hidden layers
• ReLU activation functions
• Gradient descent optimization
• Weight/bias initialization

Usage

1. Open index.html in a browser
2. Watch the visualization automatically start
3. Blue points show the target function
4. Red surface shows the network's approximation

Dependencies

• Three.js (r134+)
• WebGL-capable browser

Implementation

Core components:

• Neural network implementation
• Three.js visualization
• Training loop
• Dynamic mesh generation

License

MIT License