I help run a reading group at work. We read papers and gather to discuss them once a week. Lately we've been going through Och and Ney's "The alignment template approach to statistical machine translation" (aka, "our machine translation technique is unstoppable") and it -- both the paper, but more importantly the group -- has been pretty awesome. [I pretty much want to have the babies of all of these papers
I've participated in a few (fewer than I'd like) reading groups, and I think we've hit upon a golden technique that I'd like to share with you. First, you pick someone as the leader -- perhaps the person who suggested the paper, but definitely not the author of the paper -- and then as a group you basically walk through it page by page as the leader explains their understanding of what goes on, discussing disagreements as they arise.
And that's it. It's amazing how you can think you understand an equation (or even an entire paper) but have your understanding completely destroyed once you compare notes with others. After you get over feeling stupid, the amount of learning -- not only gaining understanding for yourself, but also explaining it well enough to help others understand -- is truly monumental, and far beyond what I ever get out of something I read on my own.
Props to mattm
for hooking it up.