Neural Network Forward Pass - Interactive 3D Visualization

This interactive 3D demo lets you build a fully configurable neural network and watch the forward pass execute layer by layer. Adjust the number of input neurons, hidden layers, neurons per layer, and output neurons using the controls panel. Choose an activation function — ReLU, Sigmoid, or Tanh — and see exactly how each neuron's activation value is computed from the weighted sum of its inputs. Every edge is colored by its weight (purple for positive, rose for negative), and every neuron lights up green as its activation is computed. Hover any element for a plain-English explanation of what it represents and why it matters.

• Configure up to 6 input neurons, 4 hidden layers, 8 neurons per layer, and 4 output neurons
• Set individual input values with sliders (0.0 to 1.0) to see how different inputs propagate
• Watch layer-by-layer activation: each layer animates in sequence showing the matrix multiply and activation step
• Randomize weights to see how the same input can produce completely different outputs
• All-layer activation panel shows every neuron's output value after the forward pass completes
• ReLU zeros out negatives, Sigmoid squashes to (0,1), Tanh squashes to (-1,1) - see the difference live

Part of the EngineersOfAI Interactive 3D collection - free visualizations covering every major concept in machine learning and AI engineering. No account required.