This paper is very different from what usually read. It’s a paper from 2010, pre-deep era. The paper is mostly about how to efficiently use high-order graphical model to do structured prediction via a cascade model.

The reason I read this paper is because I have to read it. It’s part of my qualifying exam. The way we have our qual, is the committee chair will assign you a paper outside of your research field to read; and you need to hand in a summary of the paper and also present it to the committee, and in principle the committee members can ask anything related, even slightly related.

My committee members are all nice people, however I still got a conditional pass, which was a sad story.

Anyway, I’m gonna stop here, because I do have a summary. Here is the summary

And also, if you want to read the original paper, link.

The author also have a journal version released (but not being accepted anywhere), link.