Grid size: 15×15 Learning rate η: 0.50 Neighbourhood σ: 5.0 Dataset: Visualisation:

Epoch0

Quant. Error—

Topo. Error—

Current η—

Current σ—

Kohonen SOM
BMU: argmin‖x−wᵢ‖
Update: wᵢ += η·h(dᵢ,t)·(x−wᵢ)
h = exp(−dᵢ²/2σ²) Gaussian neighbourhood.
Both η and σ decay over time.