An experiment with awesome deep learning papers

There have been several lists of deep learning papers doing the rounds. Recently Terry Taewoong Um’s list of the top 100 awesome and most cited deep learning papers caught my eye. Deep learning is an exciting area and it’s moving fast. I’d like to know what’s in those 100 papers (thankfully, we have at least looked at some of them before), and I suspect many of you would too. The problem is, at five papers per week it would take 20 weeks of The Morning Paper to cover the list! That’s too long, and even if it wasn’t, it wouldn’t match my goals for diversity of content. Ideally I’d press pause on the world for a while, go take a bunch of math classes, read through all of these papers, and then be in a strong position to understand the 2017 intake. That’s not possible either, so we need a faster plan…

This week on The Morning Paper therefore I’m trying something a little different: we’ll be working through some of the top 100 papers list in sections, covering multiple papers per day but with shorter reviews of each. Hopefully you’ll find it a good way to take on board a lot of research very quickly.

From my perspective the results are mixed – I certainly covered a lot of ground (26 deep learning papers read and reviewed over the week), but I think I discovered my current limit for paper reviews per week in the process! (It happened to coincide with a a busy work week as well). Another consequence is that the posts are longer than my usual target length, coming in at about 15 mins reading time according to my editor’s toolbar. I fear that may be too long, even in condensed review format. Let me know what you think.

One thing’s for sure, for the following week I’m going back to just one paper per day thank you very much! (And we’ll be looking at something other than deep learning too 😉 ).