Adversarial attack with respect to the Ordered Weighted L1 (OWL) norm via the ProxLogBarrier framework.
The proximal operator for the OWL norm was taken from the implementation in the following repository: https://github.com/cmazzaanthony/ccgowl
Currently the weights are pre-specified to fit the "Octagonal Selection and Clustering Algorithm for Regression" (OSCAR) regularizer, though changing this is straight-forward.