🧠 3D Neural Network Forward Pass

Every neuron computes the same formula shown in the 2D original: y = f(Σ wᵢxᵢ + b), where wᵢ are the connection weights, xᵢ the previous layer's outputs, b a per-neuron bias, and f a selectable activation function (sigmoid, ReLU, tanh or step). "Rebuild network" creates a fresh architecture — input, hidden 1, hidden 2 and output layer sizes are all adjustable — with the same random weight/bias initialization scheme as the original 2D simulator, positioned as real 3D coordinates: each layer sits on the X axis, and its neurons are arranged in a circle in the Y-Z plane, giving actual depth instead of a flat diagram. "Run forward pass" feeds random inputs through the network and computes every neuron's real activation instantly, then reveals it layer by layer with traveling pulses along the actual weighted connections — color-coded blue for positive weights and red for negative, exactly like the 2D version's connection coloring, brightness set by how strongly each neuron fired.

3D · Three.js / WebGL renderer · 60 FPS target · runs fully client-side, no install