Self-Organizing Map — Cortical Topology
Kohonen SOM learns a topology-preserving map of input data. Models how sensory cortex organizes feature representations through Hebbian-like competitive learning.
Epoch: 0 | Quantization error: --
Kohonen (1982): winner w* = argmin‖x−wᵢ‖, update wᵢ ← wᵢ + η(t)·h(‖i−i*‖,σ(t))·(x−wᵢ). Neighborhood h Gaussian, radius σ(t)→0. Left: weight vectors in data space. Right: U-matrix (neighbor distances, reveals cluster structure).