regularization (under construction)

Summary

-No Summary-

This concept is still under construction.

This concept has the prerequisites:

ridge regression (Ridge regression is an instructive example of regularization.)
LASSO (LASSO is an instructive example of regularization.)
optimization problems (Regularized learning algorithms are usually formulated as optimization problems.)
generalization (Regularization is a strategy for improving generalization performance.)

Common regularizers include:
- L_1 regularization , which encourages sparsity
- L_2 regularization , which shrinks the coefficients towards zero
- group sparsity , where multiple coefficients are encouraged to be zero or nonzero together
- Tikhonov regularization , which penalizes the sensitivity of the model's outputs to noises in the inputs
The trace norm is a regularizer for matrices which has been used in the Netflix competition.
Regularization can equivalently be viewed as adding constraints to a model, a view taken in structural risk minimization