Home

Neural Network Playground Wiki

Welcome to the documentation for Neural Network Playground, an interactive educational tool for exploring how a dense neural network learns from a single observation.

The goal of this project is to make the training process visible: forward propagation, backpropagation, gradients, weights, biases, loss evolution, and parameter updates.

Wiki Pages

• User Guide
• Network Configuration
• Training Controls
• Forward Pass
• Backpropagation
• Exported Charts
• Exported Tables
• Keyboard Shortcuts

Main Features

• Interactive neural network canvas.
• Step-by-step training visualization.
• Layer, node, and term animation modes.
• Header playback controls with adjustable speed.
• Manual weight and bias editing with custom initialization.
• Loss chart by epoch.
• Loss chart export as SVG, PNG, or PDF.
• Exportable training tables.
• Keyboard controls for faster interaction.

Recommended Learning Flow

1. Configure a small network, such as 2,2,1.
2. Select an initializer.
3. Set input and target values.
4. Use Step to inspect the computation slowly.
5. Use Epoch or Play to observe training over time.
6. Adjust playback speed when you want slower inspection or faster animation.
7. Export the loss chart or open Generate Table to inspect numerical results.