Neural network

ColorMapSOM

A Kohonen self-organizing map trained on random RGB colors. 480 neurons on a 12x40 lattice compete and cooperate until they self-arrange into a smooth, topology-preserving color field - no gradient drawn by hand, only learning.

How to read it: each cell is a 3-D RGB neuron. Because a SOM preserves input topology, neurons that sit next to each other converge to visually similar colors - so the map reads as a smooth gradient rather than noise. Hover any cell to read its learned hex value.

The trained lattice

The actual trained lattice emitted by the app (fixed seed 42, reproducible). The map learned the 24 training colors, dropping its mean quantization error from 0.072276 to 0.065589 - a 9.3% improvement. Adjacent neurons hold similar colors because a SOM preserves the topology of its input.