Lasso (Statistics)

#Math
Regularization method that tends to cause all but one parameter to be set to 0 during the process of loss minimization. Useful for reducing dimensionality of a model
See Desmos demo for why the above happens

$\displaystyle {\lambda}{1}\sum{j = 1}^{d}\lvert \beta_{j}\rvert=\lambda \lVert \beta\rVert_{1}$

This lasso term effectively minimizes the coefficients for any feature $\displaystyle \beta$
$\displaystyle J$ is the number of parameters
Effectively uses the L1 norm on each parameter
This term is added to the loss function