| layout |
|---|
default |
[[My main blog]({{ site.personal_url }}/about) has information about me.]
This site roughly follows the the Zettelkasten ideas on how to organize notes about a research field. It consists of [short summaries of papers]({{ "papers/" | relative_url }}) and [notes about concepts in the papers]({{ "notes/" | relative_url }}) related to what I am working on or that I am thinking about.
The main organizing principle in a Zettelkasten is aggressive cross-linking between different pages. This allows the structure to evolve organically as you add links and notes and it encourages you to discover new connections between papers and topics.
I originally organized my notes into a number of topics and tried to write notes about those topics. This did not work well for several reasons: it is hard to come up with a good taxonomy; it is hard to figure out where to put a paper when it is the first in some sub-topic; I am typically only reading some narrow slice of a field skewed by my interest; I am often an outsider/newcomer to the field struggling to make sense of what I am reading and struggling to find the important papers in a sea of publications.
My summaries of papers are often brief and based on a single quick skim of the paper. If you are an author and feel that I misunderstood your paper, I am happy (eager!) to correct details if you want to get in touch.
Most of the papers on this site have not been summarized because I read them before I started this site, because I have only skimmed them so far, because they are written by me or because I just wanted to remember an interesting sounding paper. This backlog of papers is growing faster than I can read papers on the backlog. Someday, I will need to come up with a plan to handle this better...
This site is the result of my need to learn a new research field where I am a complete newcomer and don't know the shape of the field: the important papers, the important research groups/people, the important concepts. In fact, I don't even know the terminology and how it is normally used.
I have had to do this several times in my career so I have some recommendations.
-
Hunt for survey articles. These may be a few years out of date, may not fit your interest/focus, etc. but they are great for getting an overview of a topic, getting the terminology (as it was used when the paper was written) and finding the important papers.
I have [a list of survey articles]({{ "notes/survey" | relative_url }}) and you can find more in places like ACM Computer Surveys or the Encyclopedia of Computer Science and the major publishers sometimes have books with names like "The Handbook of ...".
-
If you have a list of papers (e.g., some subset of the papers in this list or a list of classic papers in the field or papers from a survey), collect all the papers and then read them in the order that they were published. Reading them in this order helps you understand how the field developed, what contribution a paper made to the field, how the terminology and notation changed over time, what ideas were once important but now seem to be ignored or a bit old-fashioned. I find this to be the most satisfying and rewarding order to read papers.
(As you do this, your interests will evolve and you will find other papers and topics that you ought to read. So you will also have to hunt for papers and try to evaluate the importance of papers by following the next criteria.)
-
If you do not have a list of papers, the key thing is to create your own list.
Start with the most recent major conference in the field and skim the Related Work section of every paper looking for:
- which authors are frequently cited
- which papers are referred to as "seminal work" or similar
- which papers are often cited
- which other conferences are the important papers published at and update your list of papers.1
At this stage, you will have a list of papers that you think might be important. Follow up each of these leads to find more papers, conferences and authors that you ought to read. Try to evaluate the importance of a paper or author by looking at the citation count in Google Scholar versus the age of the paper and by looking at the related papers. Now read a few of the papers and use any new understanding to update your list.
Unfortunately, this will result in you reading the newest papers first. Oh well, it's the best you can do...
Often you will have to use a hybrid approach. You might find a [survey article]({{ "notes/survey" | relative_url }}) in ACM Computer Surveys or the Encyclopedia of Computer Science or some other book but it will always be a few years out of date. If you find one, then you can use method (1) on the papers in that survey and only have to use method (2) for more recent papers.
I like to print papers so that I can stuff them in my pocket, read them on the train, underline sections, put questions to myself in the margins, etc.
After reading each paper, make notes about the paper (maybe a bit like the notes on this site). I think it is also a good idea to try to write notes about ideas that span multiple papers both to reduce redundancy and to encourage you to identify the key concepts in a field.
Writing notes about papers and concepts forces you to think about how to explain the material – which is a great way of forcing yourself to understand it. It also forces you to practice using any new terminology you find in the papers. For example, just what is the difference between "symbolic evaluation" and "symbolic execution"? Finally, if you find yourself writing a paper in the field, then parts of your notes about a concept or a paper can maybe be copied directly into the paper. (At least, that is what the Zettelkasten advocates claim.)
As you read each paper, be sure to record enough information that you will be able to find the paper and cite it in the future. My approach is to get a BibTeX file for the paper and use a script to convert that to json that I include in the summary that I write. I believe that I could reconstruct a BibTeX file from that information but I have not yet written the script to do so. Alternatively, the doi information should be enough.
It is probably a bad idea to copy the content of this site – you will learn more by trying to write your own summaries and notes. But, by all means, copy the Jekyll framework and scripts and, if it is relevant to you, my list of interesting papers. (But maybe you should consider using a wiki instead?)
The tooling around this site is pretty simple at the moment: I have repurposed the Jekyll material from my website (which is based on Barry Clark's Jekyll Now) and I have a Python script that converts BibTeX entries to page templates. But ideally this would be much more integrated with one or more of the main research search engines like Google Scholar – perhaps as some form of overlay over the basic website.
I recently updated the [404 page]({{"404.html"| relative_url }}) based on the awesome glitch effect on the SeaHorn 404 page that is based on this glitch effect. This may hurt your eyes.
I have been slowly working on a clean rebuild of this site based on the Jekyll default site profile with the minimum number of changes to give it the important features of this site.
Footnotes
-
If you use Google Scholar to lookup papers, it can be handy to "star" papers to keep track of the interesting papers. ↩