Interactive Three.js template · AETumi.app
Aurelian Synapse — a neural network you can watch learn
Aurelian Synapse is not an animation of a neural network — it is one. A 17 → 26 → 30 → 21 → 14 multilayer perceptron with 2,237 parameters runs forward passes and backpropagation in your browser, hundreds of times per second, learning to imitate a hidden "teacher" network. Every neuron glows with its real activation and every one of the 2,146 golden weights is coloured by its real value.
What you can do
- Drag the network to orbit it 360° and see the five layers stacked in depth.
- Hover any neuron to isolate its connections and read its activation and bias.
- Click a layer header to switch its activation (ReLU, Tanh, Sigmoid, GELU, Linear) and watch the loss curve react.
- Open the ⚙ button to change the theme, speed, density, pulse size, 3D depth and zoom.
Signal pulses race along the strongest weights, the training console tracks loss, accuracy and outputs against targets, and the weight stream shows live numbers from each weight matrix.
Designed and built by AETumi Corp. Explore more Three.js and WebGL templates at aetumi.app/threejs — see also Sapphire Maelstrom, Aurum Singularity and Intelligence Router.