The number of research papers in Artificial Intelligence has reached un-manageable proportions. Conferences such as ICML, NIPS, and ICLR others are getting record amounts of paper submissions. In addition, tens of AI-related papers get uploaded to arXiv every weekday. With all these papers, it can be easy to feel lost and overwhelmed.

Like many researchers, I think I do not read enough research papers. This year, I resolved to change that, so I started an open-source GitHub repository called “Paper Notes” where I list papers that I’ve read along with my personal notes and summaries, if any. Papers without such notes are currently on my TODO radar.

After almost three months, I’m somewhat pleased with my reading progress. There are a healthy number of papers (plus notes) listed, arranged by subject matter and then further arranged by year. Not enough for me, but certainly not terrible either.

I was inspired to make this by seeing Denny Britz’s similar repository, along with Adrian Colyer’s blog. My repository is similar to Britz’s, though my aim is not to list all papers in Deep Learning, but to write down the ones that I actually plan to read at some point. (I see other repositories where people simply list Deep Learning papers without notes, which seems pretty pointless to me.) Colyer’s blog posts represent the kind of notes that I’d like to take for each paper, but I know that I can’t dedicate that much time to fine-tuning notes.

Why did I choose GitHub as the backend for my paper management, rather than something like Mendeley? First, GitHub is the default place where (pretty much) everyone in AI puts their open-source stuff: blogs, code, you name it. I’m already used to GitHub, so Mendeley would have to provide some serious benefit for me to switch over. I also don’t need to use advanced annotation and organizing materials, given that the top papers are easily searchable online (including their BibTeX references). In addition, by making my Paper Notes repository online, I can show this as evidence to others that I’m reading papers. Maybe this will even impress a few folks, and I say this only because everyone wants to be noticed in some way; that’s partly Colyer’s inspiration for his blog. So I think, on balance, it will be useful for me to keep updating this repository.