發(fā)信人: icyriver (Icyriver), 信區(qū): DataMining
標(biāo)  題: How to do Research At the MIT AI Lab
發(fā)信站: 南京大學(xué)小百合站 (Fri Jun 20 20:02:21 2003)

讀研前應(yīng)該一讀此文。做研一段時(shí)間后再讀，更是發(fā)現(xiàn)文中每處都說得如此真此。

特別emotion那部分，讓人感動(dòng)和振奮。


BBS貼效果不好。原文在網(wǎng)上可以找到。



MASSACHUSETTS INSTITUTE OF TECHNOLOGY 


ARTIFICIAL INTELLIGENCE LABORATORY


 


AI Working Paper 316 October, 1988


How to do Research


At the MIT AI Lab


by



a whole bunch of current, former, and honorary


MIT AI Lab graduate students


 


David Chapman, Editor


Version 1.3, September, 1988.


Abstract 


This document presumptuously purports to explain how to do research . We give 
heuristics that may be useful in picking up the specific skills needed for res
earch (reading, writing, programming) and for understanding and enjoying the p
rocess itself (methodology, topic and advisor selection, and emotional factors
).


Copyright 1987, 1988 by the authors.



A.I. Laboratory Working Papers are produced for internal circulation, and may 
contain information that is, for example, too preliminary or too detailed for 
formal publication. It is not intended that they should be considered papers t
o which reference can be made in the literature.


Contents

1. Introduction

2. Reading AI

3. Getting connected

4. Learning other fields

5. Notebooks

6. Writing

7. Talks

8. Programming

9. Advisors

10. The thesis

11. Research methodology

12. Emotional factors

Endnote 


1. Introduction 


What is this?

There's no guaranteed recipe for success at research . This document collects 
a lot of informal rules-of-thumb advice that may help.


Who's it for?

This document is written for new graduate students at the MIT AI Laboratory. H
owever, it may be useful to many others doing research in AI at other institut
ions. People even in other fields have found parts of it useful.


How do I use it?

It's too long to read in one sitting. It's best to browse. Most people have fo
und that it's useful to flip through the whole thing to see what's in it and t
hen to refer back to sections when they are relevant to their current research
 problems.



The document is divided roughly in halves. The first several sections talk abo
ut the concrete skills you need: reading, writing, programming, and so on. The
 later sections talk about the process of research : what it's like, how to go
 at it, how to choose an advisor and topic, and how to handle it emotionally. 
Most readers have reported that these later sections are in the long run more 
useful and interesting than the earlier ones.


Section 2 is about getting grounded in AI by reading. It points at the most im
portant journals and has some tips on how to read. 

Section 3 is about becoming a member of the AI community: getting connected to
 a network of people who will keep you up to date on what's happening and what
 you need to read. 

Section 4 is about learning about fields related to AI. You'll want to have a 
basic understanding of several of these and probably in-depth understanding of
 one or two. 

Section 5 is about keeping a research notebook. 

Section 6 is about writing papers and theses; about writing and using comments
 on drafts; and about getting published. 

Section 7 is about giving research talks. 

Section 8 is about programming. AI programming may be different from the sorts
 you're used to. 

Section 9 is about the most important choice of your graduate career, that of 
your advisor. Different advisors have different styles; this section gives som
e heuristics for finding one who will suit you. An advisor is a resource you n
eed to know how to use; this section tells you how. 

Section 10 is about theses. Your thesis, or theses, will occupy most of your t
ime during most of your graduate student career. The section gives advice on c
hoosing a topic and avoiding wasting time. 

Section 11 is on research methodology. This section mostly hasn't been written
 yet. 

Section 12 is perhaps the most important section: it's about emotional factors
 in the process of research . It tells how to deal with failure, how to set go
als, how to get unstuck, how to avoid insecurity, maintain self-esteem, and ha
ve fun in the process. 

This document is still in a state of development; we welcome contributions and
 comments. Some sections are very incomplete. [Annotations in brackets and ita
lics indicate some of the major incompletions.] We appreciate contributions; s
end your ideas and comments to Zvona@ai.ai.mit.edu.



2. Reading AI

Many researchers spend more than half their time reading. You can learn a lot 
more quickly from other people's work than from doing your own. This section t
alks about reading within AI; section 4 covers reading about other subjects.



The time to start reading is now. Once you start seriously working on your the
sis you'll have less time, and your reading will have to be more focused on th
e topic area. During your first two years, you'll mostly be doing class work a
nd getting up to speed on AI in general. For this it suffices to read textbook
s and published journal articles. (Later, you may read mostly drafts; see sect
ion 3.)


The amount of stuff you need to have read to have a solid grounding in the fie
ld may seem intimidating, but since AI is still a small field, you can in a co
uple years read a substantial fraction of the significant papers that have bee
n published. What's a little tricky is figuring out which ones those are. Ther
e are some bibliographies that are useful: for example, the syllabi of the gra
duate AI courses. The reading lists for the AI qualifying exams at other unive
rsities---particularly Stanford---are also useful, and give you a less parochi
al outlook. If you are interested in a specific subfield, go to a senior grad 
student in that subfield and ask him what are the ten most important papers an
d see if he'll lend you copies to Xerox. Recently there have been appearing a 
lot of good edited collections of papers from a subfield, published particular
ly by Morgan-Kauffman.


The AI lab has three internal publication series, the Working Papers, Memos, a
nd Technical Reports, in increasing order of formality. They are available on 
racks in the eighth floor play room. Go back through the last couple years of 
them and snag copies of any that look remotely interesting. Besides the fact t
hat a lot of them are significant papers, it's politically very important to b
e current on what people in your lab are doing.


There's a whole bunch of journals about AI, and you could spend all your time 
reading them. Fortunately, only a few are worth looking at. The principal jour
nal for central-systems stuff is Artificial Intelligence, also referred to as 
``the Journal of Artificial Intelligence'', or ``AIJ''. Most of the really imp
ortant papers in AI eventually make it into AIJ, so it's worth scanning throug
h back issues every year or so; but a lot of what it prints is really boring. 
Computational Intelligence is a new competitor that's worth checking out. Cogn
itive Science also prints a fair number of significant AI papers. Machine Lear
ning is the main source on what it says. IEEE PAMI is probably the best establ
ished vision journal; two or three interesting papers per issue. The Internati
onal Journal of Computer Vision (IJCV) is new and so far has been interesting.
 Papers in Robotics Research are mostly on dynamics; sometimes it also has a l
andmark AIish robotics paper. IEEE Robotics and Automation has occasional good
 papers.


It's worth going to your computer science library ( MIT 's is on the first flo
or of Tech Square) every year or so and flipping through the last year's worth
 of AI technical reports from other universities and reading the ones that loo
k interesting.


Reading papers is a skill that takes practice. You can't afford to read in ful
l all the papers that come to you. There are three phases to reading one. The 
first is to see if there's anything of interest in it at all. AI papers have a
bstracts, which are supposed to tell you what's in them, but frequently don't;
 so you have to jump about, reading a bit here or there, to find out what the 
authors actually did. The table of contents, conclusion section, and introduct
ion are good places to look. If all else fails, you may have to actually flip 
through the whole thing. Once you've figured out what in general the paper is 
about and what the claimed contribution is, you can decide whether or not to g
o on to the second phase, which is to find the part of the paper that has the 
good stuff. Most fifteen page papers could profitably be rewritten as one-page
 papers; you need to look for the page that has the exciting stuff. Often this
 is hidden somewhere unlikely. What the author finds interesting about his wor
k may not be interesting to you, and vice versa. Finally, you may go back and 
read the whole paper through if it seems worthwhile.


Read with a question in mind. ``How can I use this?'' ``Does this really do wh
at the author claims?'' ``What if...?'' Understanding what result has been pre
sented is not the same as understanding the paper. Most of the understanding i
s in figuring out the motivations, the choices the authors made (many of them 
implicit), whether the assumptions and formalizations are realistic, what dire
ctions the work suggests, the problems lying just over the horizon, the patter
ns of difficulty that keep coming up in the author's research program, the pol
itical points the paper may be aimed at, and so forth.


It's a good idea to tie your reading and programming together. If you are inte
rested in an area and read a few papers about it, try implementing toy version
s of the programs being described. This gives you a more concrete understandin
g.


Most AI labs are sadly inbred and insular; people often mostly read and cite w
ork done only at their own school. Other institutions have different ways of t
hinking about problems, and it is worth reading, taking seriously, and referen
cing their work, even if you think you know what's wrong with them.


Often someone will hand you a book or paper and exclaim that you should read i
t because it's (a) the most brilliant thing ever written and/or (b) precisely 
applicable to your own research . Usually when you actually read it, you will 
find it not particularly brilliant and only vaguely applicable. This can be pe
rplexing. ``Is there something wrong with me? Am I missing something?'' The tr
uth, most often, is that reading the book or paper in question has, more or le
ss by chance, made your friend think something useful about your research topi
c by catalyzing a line of thought that was already forming in their head.


3. Getting connected

After the first year or two, you'll have some idea of what subfield you are go
ing to be working in. At this point---or even earlier---it's important to get 
plugged into the Secret Paper Passing Network. This informal organization is w
here all the action in AI really is. Trend-setting work eventually turns into