|
I'm a brazillian phd student that wants to work on machine learning and nlp, but is finding it a bit hard to think and do research at the level I want (international, etc). I think one of the problems I have is that I have no idea which problems to study, so I spend a lot of time reading papers trying to squeeze an ounce of possible future work from it, and then find out that it's either hard or someone has already done it or it's useless and uninteresting. I suppose the best way to overcome this is to try to get more inside the actual community, but it's not easy, for someone without physical contacts. I went to NIPS 2009 without a paper, and found it hard to strike up a conversation with anyone. It feels as though there's a chicken-and-egg problem: to do good research you need to have an intuitive feel for what are the good open problems (that are solvable by someone new) and collaborators, and to get this feel and collaborators it really helps to have done research for a while. My question, then, is: what are the common ways used to break out of this cycle? Does anyone have specific advice to offer on this question? Who could be a good advisor, who will advise me remotely?
This question is marked "community wiki".
|
|
[sorry I've added a bunch of ideas, so my thoughts aren't organized.] If you don't get to communicate with other people, and don't get good advice from your colleagues, it's hard to get a sense what people care about. You need a good advisor. I suspect there are many good associate professors (if not full professors) like this:
If you already have funding, I think you can find someone for a remote collaboration. I assume you will do all the coding and experimentation and are respectful of the advice you get during the limited time you receive. When picking an idea, make sure to ask a handful of different people whose opinion you respect (via email) for their feedback. Make a list of advisors with whom you would like to collaborate. To what extent is it possible for you to study abroad and/or switch your location of Ph.D. study? Usually people work with whomsoever is around them. That kinda sucks. I've formed collaborations by just emailing people whose work I think is cool, and telling them I want to work with them. I'm not shy to email ten people in the hope that one will give me a positive response. I think you could try this too. I would pick eminent professors. I would also pick grad students who look like they are going to make a big impact, especially those that work with eminent profs. Regarding cold emails, it's unusual, but I've done it. Getting a referral is better. Lastly, I think your answers on this site speak immensely to your ability. I totally agree here--a good advisor will not only tell you what's interesting, but warn you quickly if you're going down an unproductive path. I've wasted huge amounts of time in research going down a path only to find it's either unproductive or doesn't address my original problem: a good advisor will see those problems from much further away.
(Jul 06 '10 at 23:48)
sbirch
|
|
I think the advisor is very important, but most of the ones who are good/famous researchers are usually very busy and you'll find yourself on your own again. I had a good experience doing an internship with Google, and colleagues that went to research labs of other companies such as Microsoft and Yahoo also got good research ideas from their internships. There, the problems will run to you rather than the other way around, plus it will give you time to survey literature on a given topic and see what are the problems with the state-of-the-art. So my advice would be to go for an internship with a research lab or a research oriented company focused on, or interested in, machine learning and nlp projects.
This answer is marked "community wiki".
And how does one go about applying to one of these labs? What is usually expected that the person has already done? What are the big ones (I can think of Google, Yahoo, Microsoft research, IBM's T J Watson, NEC, Bell labs; are there other significant ones?)?
(Jul 07 '10 at 07:18)
Alexandre Passos ♦
1
You can contact the labs directly with your current CV (where you should emphasise your interest in/experience with machine learning projects). Lots of MSc and PhD students typically apply for internships with them. The lab will typically set a phone interview with you and their HR that will try to match you with a technical person with background/interests similar to yours. They might choose to give you some small puzzles and check your coding skills to some level. I know about the european offices/research labs of Google, Yahoo and Microsoft and they are all excellent. For Google the Zuerich office has great ML projects, for Yahoo the Barcelona research lab is excellent and for Microsoft they have a great group in Cambridge. Of course the main projects are usually in US, but since Europe has a strong hold in ML projects as well, its up to you to choose where you want to go.
(Jul 08 '10 at 03:45)
Georgiana Ifrim
|
|
Hi Alexandre,
This answer is marked "community wiki".
|
|
Maybe apply to spend 6-12 months in an American university, maybe just as a masters student. That will help you build connections with both students and professors you could stay in touch with in the future and stay updated about the latest happenings in the field.
This answer is marked "community wiki".
|
|
Well, a really good way is fairly obvious: get involved in development of an (already existing) open-source NLP/ML project. At first, you'll probably be assigned relatively menial tasks like writing documentation, but you could make experienced contacts fast. Then at the same time always be on the lookout for opportunities to publish, even if it's mediocre and not "good" research -- everyone has to start somewhere. Also, you could look into some of the contests at kaggle.com. They have a machine learning implementation contest going on where the first prize is $500 and a coauthorship.
This answer is marked "community wiki".
2
I think this is poor advice. Alexandre wants to have an impact in academia. Open source projects rarely have an opportunity in academia, it's usually the other way around. Your advice about be on the lookout to publish, even it's it not good research, points Alexandre in the wrong direction. He should focus on a high-impact, home run style publication in top conferences. To do that, you need a good advisor.
(Jul 06 '10 at 18:04)
Joseph Turian ♦♦
2
If you look at Alexandre's webpage, he has already published, and he says it hasn't worked. My feeling is that he is picking poor topics. If you hack on someone's open-source code project, you might get pushed into helping them with the engineering, rather than doing original research. It's better to get involved in an existing research initiative than an existing code initiative. There are many good advisors. It's not just Charniak.
(Jul 06 '10 at 20:00)
Joseph Turian ♦♦
|
What I do: closely follow some blogs and the main conferences in the area, reading a few papers from every conference when they're released and later reading others as my interested is pointed towards them.