A Connections to other problems In the Introduction, we claimed that several well-known formulations for robust learning can be recovered as approximations of (PII)

Neural Information Processing Systems 

More complex choices of distributions lead to other robust learning methods involving random removal of patches from the image (e.g.,