L1 regularization
Also: L1, lasso, lasso regression, L1 penalty
A penalty λ·Σ|wᵢ| added to the loss. Because its constraint region has sharp corners on the axes, the optimum often lands where some weights are exactly zero, so L1 doubles as automatic feature selection.
L_total = L_data + λ Σ |wᵢ|
Example
For w = (3, −4) and λ = 0.1 the penalty is 0.1 · (3 + 4) = 0.7; as λ grows, the weight on a weak feature typically snaps to 0 rather than just getting small.