L2 regularization
Also: L2, ridge, ridge regression, weight decay, Tikhonov regularization
A penalty λ·Σwᵢ² added to the loss, which pulls every weight smoothly toward the origin and spreads importance across correlated features. With plain gradient descent it is equivalent to weight decay: each step multiplies the weights by (1 − 2ηλ).
L_total = L_data + λ Σ wᵢ²
Example
For w = (3, −4) and λ = 0.1 the penalty is 0.1 · (9 + 16) = 2.5; the big weight is penalized most, so both weights shrink but neither reaches 0.