|
|
|
| New insights into one-norm solvers from the Pareto curve | |
|
Next: Solution paths
Up: Hennenfent et al.: Pareto
Previous: Pareto curve
To illustrate the usefulness of the Pareto curve, we
compare IST, ISTc, SPG
, and IRLS on a noise-free problem and
compute a solution of BP
for
, i.e., BP
. This
case is especially challenging for solvers that attack
QP
--e.g., IST, ISTc and IRLS--because the corresponding
solution can only be attained in the limit as
.
We construct a benchmark problem that is typically used in the
compressed sensing literature (Donoho et al., 2006). The matrix
is taken to have Gaussian independent and
identically-distributed entries; a sparse solution
is
randomly generated, and the ``observations''
are computed
according to equation 1.
Subsections
2008-03-27