به نام خدا
For the past two years, or a little more, I have been following the latest research published on arXiv. I also check the proceedings of relevant conferences, but arXiv is my most important resources for following what is going on in the world of AI.
I am subscribed to the RSS feeds of some of the relevant arXiv sections (cs.CV, cs.LG, cs.NE, cs.CL, etc.). So all new entries are available in my feed reader of choice (Feedly). At first I didn't have a schedule to check the new entries, and sometimes entries got removed after 30 days. However, for the past few months, I follow a plan that helps me keep track of all the latest research. I have arXiv Saturdays, when I check all of the papers added to arXiv during the past week. Why Saturday? Because that's the only day no new entries are added to arXiv. This also helps with amortizing the time needed to check all these papers, since each week a large number of new papers are added to arXiv, for example today that I checked Feedly, there were 470 new entries!
As I scroll through the new entries, the first filter is the title of the paper. Judging by the title I can filter out most of the papers. If I find the title relevant to my own and some of my friends (more on the friends later) research, I move to read the paper's abstract. I not only check out the papers relevant to my own research interests, but I also check some that are relevant to the research interests of a number of my friends. If I find them relevant, I email the link to the article to them. I know that they now hate my emails, and I'm almost sure that they don't read most of them :D Anyway, it helps me get to know what is happening outside of my research interests. In addition to the abstracts, I also check the list of authors. I know that some of the authors publish on the same topics that I'm working on most of the time. If a paper passes through these papers, then it gets added to an special "list" on a Trello board that I call "Reading List". Papers stay on the "To Skim" list until I get back to them and then move them to the lists I have created for each project or topic. You can say that this reading list is quite big by now. Given that I have created this Trello board just months ago, I have started to feel that a Trello board might not have been the best choice for managing these number of items. However, so far I haven't found a replacement. In Trello, a card is created for each paper and it contains a link to the paper, sometimes a link to an implementation of the paper, and also a short summary or note I write about the paper from time to time. I also add labels to each card, which indicate whether I have already read the paper, is it work-related or not, what general topic does it belong to, and some other labels. However, you can't link cards to each other. As a result, you can't have a good ordering between a subset of papers. The problem is that to have proper ordering between papers, you need to have a graph, a simple list won't make it. That's why I still have problem ordering papers according to some features, around a central topic or research question.
To manage the actual articles, which I finally get to read, I use Mendeley. It is indeed a great service, it has a very good cross platform application and its document reader is quite functional. I am one of those who highlights papers heavily and in fact I have devised a color code for my highlights.
But in addition to reading the papers and highlighting them, I need somewhere to write other notes and sometimes try to work out the equations for myself once more. For that I use OneNote, one page for each paper.
The combination of Mendeley and OneNote works fine for the time being, although my cloud storage provided by Mendeley is filling up quickly.
Since I'm just a beginner, most of my time is spent reading other researchers' published work, but even now I feel that this process is not quite working. The discovery part is fine I think. I check out everything on arXiv. I also follow a large number of researchers and PhD students on Twitter and this stream is quite useful in staying on top of the best and latest research. The reading and taking note part is also a working combination. However, the managing and scheduling side of things is not close to being called "working". I still think that I need to find a better tool or combination of tools for managing the comparably large volume of research papers and schedule reading them. But all these stuff aside, it is really becoming very hard keeping up with the latest research. Checking out weekly papers on arXiv is a very time-consuming activity by itself, now imagine the list of accepted papers to a conference getting published! A true paper overflow.
How have you set up your research consumption workflow? How do you manage and schedule the ever-growing list of papers?comments powered by Disqus