Jordan Matelsky
365Papers

epilogue — Epilogue

Read on 20 August 2018

What Have We Learned

Earlier today, I posted a discussion of my 365th of 365 papers.

I read one paper every day for 365 days since August 22, 2017, and today I am a year older and 365 papers smarter.

Well… Okay. I didn’t read one paper every day. But I averaged one paper every day, and that’s really really close. More on that in a second.

This was a huge undertaking, and I’ve commented not too infrequently about how it was, at times, very unpleasant:

Then why did you do this thing?

Engaging with scientific literature is hard. But despite this cost (both in time and money), inaccessibility (both vocabulary-wise and file-access-wise), and lag-time, the majority of scientific communication is still performed via peer-reviewed papers. $h$-index is still a ubiquitous measure of an academic’s impact. Number of publications is still a metric for a young academic’s career growth.

So I wanted to broaden my knowledge base, but I also wanted to find the cracks in the system where scientific knowledge trickles through despite the limitations of the publishing process. That includes social media ecosystems like scicomm-Twitter, which act as powerful conduits for scientific collaboration and discussion. My Neuro-Twitter idols are DOCTOR Caitlin Vander Weele (creator and publisher of Interstellate) and Adam Calhoun, along with all of the authors who choose to bring their narrative out of the dusty dialect of publication and into the lingua franca of public storytelling.

So many of the papers I read this year were the result of seeing a paper posted by the authors on Twitter. I tried to tag the tweet and author whenever possible (and many of my posts start with that embedded tweet) — sincerest thanks to you.

But why did you write a discussion about every paper?

I needed some way to keep myself on track, or I was bound to try to cheat. I can fudge timestamps on a blogpost, but I can’t fudge timestamps on a tweet. Sorry to everyone who had to slog through a #365papers-laden timeline for a year, but you enabled me to do this, so also thanks! Hopefully you found an interesting paper or two on my timeline, as well.

Oh, and this:

Tips & Suggestions

A surprising number of folks have asked me how they can get started doing this as well — most surprising because I didn’t know so many people even knew I was doing this!

My suggestion is… well… “don’t.” 365 is a big number, and, frankly, I don’t think I absorbed enough about each paper to do them justice. 52 papers would have been a better number, perhaps — and would have given me a week to better curate, chew on, and digest each paper properly.

But if you must, then my suggestion is to front-load a bit. I never let myself fall behind — I posted a paper every day — but I often had two or three papers in reserve. It was particularly nice when, at 11pm, I found myself at a bar, wobbly, and paperless. Have a backlog. Once you break your streak a single time, you lose a lot of the power that the project holds over you.

If you’re new to binge-reading academic manuscripts, I would also highly recommend reading about a week’s worth of papers ahead of time before you start, so that on the first day of posting, you have a week of backlog. Still try to keep up and read another every day — but this type of reading is different from normal reading, and it’s a new skill you’ll have to learn. You can expect the first few papers you read to take much longer than you expect.

My Tools

  • Paper organization. For a while, I was using Mendeley to organize papers, and then I tried ReadCube, and plenty of others. It just didn’t really do it for me, but I can see the draw. Instead, I just dropped a PDF link into my Todoist in a #papers label; I always knew where my backlog of interesting papers was.
  • Writing posts. I wrote my own code leveraging GitHub Pages for writing posts in Markdown and posting them; there’s more explanation here if you want to try it out yourself, or if you want to steal my website style.
  • Reading. I tried a few PDF readers — they’re all poo. The best experience is printing it out and reading with a pen and paper; second best is a reflowed-text HTML page online; and then comes the ugly old PDF format. More on that in a sec.

Passing thoughts from a year of reading and writing

We need a better way to know if the contents of a paper are refuted or out of date. This is one of the great benefits of having good mentors: They can point you to current research. Particularly when I was out-of-domain (reading paleontology or marine biology papers), I couldn’t be sure if what I was reading was the true opinion of the field. Thank you to domain experts on Twitter who helped me feel my way around a new field!

Pre-prints are good way of pulse-checking a field, but they very much exist on a continuum of “basically publication-ready” all the way to pseudoscientific garbage. That being said, peer-reviewed papers’ topics have often aged poorly between the times when the paper was written and when it was published, and pre-prints are basically all you’ve got in fast-changing fields. Example: In light of arxiv, no one really writes neural net papers anymore unless something very big has happened.

Access — and inaccessibility — is a huge deal. I can’t recommend Unpaywall highly enough. It’s a tool that finds free PDF versions (legal!) of papers whenever they’re available. But when papers weren’t available in open-access places, I used my Johns Hopkins affiliation to download them. But… Yikes. I’m not the first one to talk about how awful this is, but I wish there were a solution to this that didn’t involve authors absorbing the enormous cost of open access.

PDFs suck. I understand why this is the standard we’ve settled on, but it’s garbage. Someone needs to fix this.

FAQs

  • Didn’t you get bored of reading? Ha. Yep.
  • What was the best paper you read? Don’t make me choose! But this one is definitely high up on the list.
  • What’s your [favorite, fastest, most exciting] type of paper to read? Connectomics papers are so exciting! Those are definitely my favorite. But I also get super excited about revolutionary machine learning papers. Everything moves so fast in ML. As for fastest… Well, my guilty confession is that picking off arxiv AI or dataset papers was basically my freebie easy-mode option for days when I really wasn’t feeling it.
  • What was your least favorite paper? I won’t point to any in particular, but the worst thing ever is academic jargonspeak. “We identify a novel mechanism by which it is demonstrated herein that [abbreviation-alphabet-soup], [citation] [citation] [citation], is downregulated by [alphabet soup] streams through which [irrelevant self-aggrandizing self-citation] is suggested.” …whaaaaat? Every time I selected a paper that sounded like that, I wound up kicking myself. It took forever to wade through that garbage. The best papers were the ones that spoke like a human telling a story to another human. When I write papers now, I use this guide by Konrad Körding et al.
  • How long does it take you to read a paper? Even the really basic, simple papers take 45 minutes or more. Usually I spend 1 to 2 hours minimum on a paper: On papers about which I’m very excited, I’ve spent at least six hours mulling over the ideas. And, of course, that doesn’t include time I spent digesting or writing. I assume that number is an underestimate.
  • Are you going to keep going after the end of the year? Hell no. But I absolutely gained an enormous amount of reading efficiency, and that’s neat. And — not for nothin’, but my job is all research. I read plenty of papers outside of #365papers, and that’s certainly not going to stop.
Comments?
@j6m8