How do you self-study?

So I want to read up on, say, natural language processing now. I have access to a million different blogs on the topic, and as many journals and papers on the topic… basically whatever’s on the Internet, apart from the stuff in the IEEE and ACM databases.

I do not have any tasks at hand in the field that I can pick and study what will help me get the task done. I just want to know more, both breadth-wise and depth-wise.

Breadth-wise seems easy as there are enough sites that give you a bird’s eye view of the whole thing.

It is depth that is challenging.

For starters, how do I go about the whole deal in an organized and efficient manner?

How do I sift through the mountains of published papers, most of which use terminologies and math that at first glance seem alien to me? It wouldn’t be as painful if Googling for those terminologies yielded definitions instead of more papers on the same, but that seems to me to be too much to expect.

How do I know what is relevant and what is not? I don’t think bruteforcing through loads of documents hoping whatever I read comes of use somewhere is the best thing to do. More so when half of it doesn’t yet make sense to me.

And what would be a good way to get a more accurate view of the breadth?

As a bonus, how do you conduct research in your spare time?

Advice on this, especially with respect to machine learning and natural language processing are solicited.


About wanderlust

just your average books-and-music person who wants to change the world.
This entry was posted in analysis, geek and tagged , , , , , , . Bookmark the permalink.

9 Responses to How do you self-study?

  1. Tuna Fish says:

    Use Scopus, with the Citing and Cited-by feature in papers, Gives more hold than random google searching…
    Start off with review papers, Then pick the papers from the parts that interest you. Supplement it with books (not the net) for better understanding…
    But then however much you read, it can never equal hands-on experience, IMO…

  2. theG says:

    only advice. Do NOT try to take shortcuts. Pick a paper, if you do not understand it (which you will NOT, because if you do, either you are very experienced, or you did it totally wrong), start checking the history of the paper. Read the references, cross reference.

    There is not shortcut to the math. Get hold of the relevant books., understand the math, and then try to apply it. Check out ACM and IEEE, see who is involved in a lot research in the area you are interested in. See what all papers are written (generally there will b elots of papers by the same author, which is a good sign). But you can never try to do things too fast, try to take shortcuts. Do the brute force to start with, but you will figure out how it works pretty soon. Best of luck!

  3. Logik says:

    Whatever theG said…google scholar might be of help there.
    It’d be best if you could separate the math bit of it, and deal with it,in the preliminary stages. Because once you pick up a paper that is often cited in a field, chances are that you’ll see that math popping up in a host of other places.
    Divide the set of papers in a time and importance based hierarchy. Time – year of publishing. Importance – ( author, topic, depth, no. of citations).
    With respect to computing, many papers of the 80’s are not yet obsolete and give a jump-start with the math. [ and were written by now-considered-gods ].

    Personal experience with Computer Vision, and Machine Learning. NLP – no clue

  4. ego says:

    Based on my limited experience, and the advice given by experts, here are my two cents:

    Use google scholar. It’s pretty decent. Use the search string to match the area of your interest.

    I click on the recent work. Two reasons why I do this.If the paper is interesting/relevant, and there’s something deeper that I need to know about, I can always look at the bibliography section. Most of the ACM/IEEE papers that way cite things pretty well.Secondly, you can keep track of who are currently active in this area.

    Now, once you’ve chosen the paper, how do you go about deciding whether it’s worth a read or not. I _used to_ read the whole thing, and then say, “oh! tall claims, but no actual benefits!”. That’s where Paul McKenney’s advice came in handy. Simple rules, but they help a lot.
    1) Read the abstract. If it’s interesting, but some terms are confusing, look them up. If still interesting/relevant, go to step 2.
    2) Read the conclusion. (Seriously! this helps). Ensure that what claims the authors made, they actually lived upto it. It passes this test, go to step3.
    3) Look at the graphs. Usually, you can determine based on graphs the improvement the authors seem to have made with their method or algorithm. Look at the units. When you’re considering absolute results, units make a lot of difference 🙂
    4) Now read the paper, top to bottom. Keep a note book handy to note down doubts, which you need to refer later.
    5) If you liked the paper, but need to know some more info, see the relevant citations and read those.

    This worked for me. Hope it helps.

  5. Abi says:

    If you want to learn stuff in a new (sub)field, introductory books are the way to go — preferably the dead tree versions. By the time you are done with the book(s), you will not only have the breadth covered, you will also have the acquaintance/familiarity with field’s deep, central ideas and the names of people behind those ideas.

    Then you go drilling for depth, by going to the source — research papers presenting and discussing these ideas.

  6. sindhu says:

    We have AI as a course subject in our 5th semester and recently i was required to take up NLP for the class..yes teach them NLP when I did not know anything myself. My best bet was Elaine Rich’s Artifical Intelligence book, which is pretty precise but i scoured a lot of info off googling, esp from IIT KGP site, because they had NLP and AI as course subjects too. You may see relevant links here, here, here and here.

    as for self study, having an object is most important, if you want to study NLP, take a textbook say the one from elaine rich in our case, study the basics first, that will give you a better direction to proceed in.

  7. karthik says:


    Then references to papers in the textbook, which are usually old papers.

    Seems to be working for me thus far. {shrug}

  8. the Monk says:

    Go for depth only after you have understood what is going on in terms of breadth. In other words, read about the whole thing, pick an area you like and *then* go for depth. Also, don’t google for papers, it’s always better to ask somebody who’s in a research institute to search for papers, as they will have direct access to many journals. For instance, ask your friends in IISc (if any). Also, try and talk to some professor in the area as to what kind of mathematical background you’ll require in order to understand these papers well. While you’re at it, you might as well ask him about sources for the breadth part of it as well. This generally works: talk to experts.

    There is no shortcut around the math if you’re actually thinking of doing research in your spare time, but in the initial stages you might want to get a qualitative understanding first and then go for complete understanding.

  9. Shreevatsa says:

    Find someone to talk to. Once you’ve overcome the notation/terminology barrier (and that’s the hardest, I find…) you’ll be able to use your common sense to find relevant papers, references, and so on (through Google Scholar, or the dozens of other means).
    If you can’t find someone, particularly valuable for jumping over the “what does this mean?” barrier are survey articles, blog posts by the researchers (ok, whether such things exist depends on the community :-)), and lecture notes of courses, etc. (Of course, if the material is old enough to be in an actual textbook, that’s ideal.)

    Once you know what things mean, there are strategies to extracting content out of a paper, but that isn’t *that* hard, usually…

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s