This module covers shrinkage. Shrinkage is an approach that helps sparsify our coefficients vector. One way to do so is ridge regression, which penalizes larger coefficients using a squared ell2 norm, thus shrinking them toward zero. The LASSO algorithm uses an ell1 for penalization; there are multiple algorithms for implementing LASSO efficiently. We then describe how to estimate a sparse coefficients vector using the Sudocodes algorithm from 2006. Under assumptions of sparsity and noiseless measurements, Sudocodes offer a very efficient solver for linear inverse problems.
