| Ray's profile点一盏心灯PhotosBlogLists |  Tools  Help |
|
|
March 02 我作RESEARCH的几点体会zz from mitbbs我作RESEARCH的几点体会 by zuyeye 发信站: BBS 未名空间站 (Mon Feb 25 14:34:20 2008) 完全个人观点，不见得试用于所有人。 第一不要太迷恋TECHNICAL的东西。从自己和周围一些朋友的例子，我发现很多中国学 生对TECHNICAL的东西特别感兴趣。什么新的东西都想学，不管究竟对自己的RESEARCH 有没有帮助。所以很多中国学生的通病是，前两年上基础课的时候都很优秀，经常是全 A。但到了三四年级时，进入RESEARCH很慢。很多人的毕业论文都是匆忙上阵，甚至迟 迟不能毕业。当然我并不是说TECHNICAL的东西不重要。博士课前两年是积累一些 GENERAL TECHNICAL SKILLS很重要。剩下的时间主要集中在开始RESEARCH和学习与自己 RESEARCH相关的TECHNICAL SKILLS。 第二要在三年级就开始选定一到两个具体的TOPIC。我觉得从什么报纸，杂志或网站上 找灵感都是瞎掰。如果自己不是JOHN NASH那样的奇才，老老实实从文献里找灵感。看 看自己的导师在作什么，看看自己领域的大牛(如自己领域TOP JOURNAL的EDITOR)在作 什么。作研究都是大牛挖坑，小牛灌水。对于JUNIOR RESEARCHERS，我感觉最有效的研 究方法就是紧跟大牛挖坑的步伐。认准一个题目后，静下心好好清理一下LITERATURE的 脉络。不要朝三暮四，这山望着那山高，整天试图去追逐所谓的HOT TOPIC。 整理文献时先作一个LITERATURE TREE：谁提出了这个TOPIC，LITERATURE里都从什么方 向研究了这个题目，它们的关系是什么。对每个分支只包括最重要的1到两个文章。把 大牛和中牛抓住就行了，剩下的小鱼小虾就算了。参阅一些LITERATURE REVIEW的文章 ，如果有的话。然后对经典的文章，如提出TOPIC的文章要进行复制。每个公式，每个 结果都试图从新作一边。如果有问题可以向原作者请教。我感觉他们大部份还是很NICE 的。复制的工作量其实很大，但也可以学到很多东西。很多东西不靠亲自作是无法体会 的。看看自己的导师和领域的大牛最近的研究在LITERATURE中的定位。复制自己感兴趣 的文章。 有了这些准备工作之后就可以进行自己的RESEARCH了。有没有什么办法把导师或大牛的 文章与其他研究联系起来，有没有什么EMPIRICAL STUDY可以做来支持导师和大牛的论 点，这些都是不错的研究方向。另外，一般文章中都会提出一些自己的不足和未来的研 究方向。这些都是不错的研究题材。我这里一直强调要跟大牛走。说白了，学术界其实 就是被这些人控制着。大家都说中国有学霸，学阀。其实什么地方都一样。论文就是观 点之争，不拉党结派怎么在江湖上混啊。对刚出道的人，最好还是厚道点。多作点 CONSTRUCTIVE的工作，少批评。我前面给的都是这方面的例子。江湖险恶，只是很多人 在作学生的时候不知道罢了。比如说JOURNAL EDITOR的权力其实很大。他们很清楚什么 人持有什么观点。他们喜欢的东西就发给持相同观点的审稿人，不喜欢的就发给不同观 点的审稿人。结果可想而知。以前的导师不久前让我审一篇稿子。作者声称发现数据支 持我导师一篇很重要的JPE文章。看了看文章的统计方法没什么大问题，我当然举双手 双脚赞成发表了。 当然也可以指出大牛文章的不足。但是不能为批评而批评。能找出改进方法解决问题的 文章才是好文章。比如可以这样写，我复制了A的模型，发现有什么地方和数据不符， 原因是什么。我对原模型作了一些改动后，现在模型和数据MATCH了。写作过程一定要 注意，多POSITIVE，少NEGATIVE。如在指出A模型和数据的冲突时，不要过多强调这种 冲突可能引起的问题。而应该强调解决问题后带来的好处。这样其实已经是对原作者的 间接批评了。如果A是领域大牛或作EDITOR，这篇文章十有八九能通过。 想跟上大牛步伐，看JOURNAL ARTICLES是不够的。这些JOURNAL上的东西至少是3-4年前 的东西了。要多留意WORKING PAPER。列出自己领域20大牛人。经常看看他们在干什么 ，和自己目前的RESEARCH有什么联系。参加一些WORKING PAPER的EMAILING LIST，比如 NBER等。常看看TOP10 DEPARTMENT本领域的SEMINAR LIST。我想再强调一次，这样作的 目的并不是要去追逐本领域的HOT TOPIC，而是看看别人的研究和自己目前的RESEARCH 有什么联系，能不能给自己点灵感。至少在自己CAREER的前3-4年把RESEARCH INTEREST 限定在两个方向，每个方向整出2篇像样的文章，再考虑换方向。少而精要比泛泛涉猎 几个不同方向更容易引起别人注意。 就先谈这么多，以后有时间了再补充。 另外，要专心作RESEARCH就要减少在MITBBS的灌水时间。每天减少一小时灌水时间，一 年能多读多少篇论文啊。没有BBS北美广大WSN能平均提前1年毕业。：-) December 11 what is the deep question of Computer Science? (2)Why Jeannette and other professionals emphasize the TWO questions? One can view the whole path from the historic perspective.
1) What is intelligence?
language: how to talk with computers?
If one can talk with computer without awareness of his facing a machine, then the intelligence come out.
After Church and Turing published their thesis in 30s, and some actual computers were put into real work in 40-50s, it came a natural problem: how to interact with computer? How to teach a computer to understand instructions? In those early days, a computer was really manually operated by technical workers, and the very early "programs" were very hard to write, machine-dependent, and even one could write, other could not easily understand. The first attack to this problem at IBM brought about the language later well known as FORTRAN. Its inventor Backus later created the famous Backus-Naur Form (BNF). The algol60 language was later defined in BNF, and it became the root for many later languages such as Pascal and C. Newell and Simon proposed an idea to manipulate symbolic processing in mid-50s, and was created by McCarthy, i.e. the LISP language that was and now still is widely used in AI. Artificial Intelligence was first coined by McCarthy.
Language is one big problem in computer science, and at that early stage, was studied by liguists, psychologists, mathematicians and logicians. why this problem was important? Turing propsed a test to judge the intelligence of machine: one is talking with one agent which is sometimes a human and sometimes replaced by a computer. If he couldnot tell when he is facing computer or not, the computer appears some kind of intelligence. December 09 the words of HammingHamming once gave a talk about research.
If he told something, that is, one should, do world class research, the kind of work that could be considered Nobel Prize.
A short time at Los Alamos National Lab, later almost 30 years at Bell Lab, Hamming worked with great scientists such as Shannon. He saw a lot world class job done and did great work himself. In his years, it's not even clear what computer science is about, while his coding theory played an important role in later development of CS.
But why you bother? To creat the greatest impact, as he implied in his talk. November 27 what is the deep question of Computer Science?Last week, Jeannette sent an email to the whole school of computer science, posting a discussion on what is the deep question of Computer Science. She, later James Morris and Peter Lee mentions two possible essential questions:
1) what is intelligence?
2) what is complexity?
James and Peter also emphasize CS is not merely programming, but far beyond! Not accidentally, they also highlight the impact of computer science should make in the society, which is much similar to what I am moved about at MSRA.
The computer science should not be a discripline that only tells people how to be rich or how to gain fame! It should have deep insight into the understanding of some aspect of our nature. November 26 some links about graduate survivalJune 20 ubuntu 6.06 installed successfully!前几天rp一边休息凉快去了，导致连刻了6张ubuntu 6.06的光盘，desktop, alternative都刻了，没有一张可以成功安装。甚至连别人试成功过的安装盘，到了我的电脑上，就是不行。
症状是用desktop在mount root file system时无法mount，hdc(光驱) io error。alternative盘在detect & mount CD时无法mount。
这个错误估计是6.06 ide驱动程序与硬件兼容性的问题，因为5.10可以成功安装。
今天rp休假回来了，ubuntu 6.06 LTS 安装成功！而且xubuntu也可以成功安装了。分享一下经验：
硬件配置：
IBM整机，CPU intel P4 2.4G, 1G memory, IBM 40G ATA（vendor是）,17 viewsonic LCD，光驱接在IDE 3号口上。
BIOS设置：
video memory所有选项都设成最大（4092K, 256M），原来default的设置比较小，导致安装后无法显示图形界面或者分辨率无法调到最高。
安装选项：
用光盘（desktop或alternative均可）启动，出现ubuntu安装选择界面后按F6然后在添加选项ide=nodma。后面就可以按照提示正常安装了。
分辨率的设置：
如果安装成功后，无法显示高的分辨率(1280*1024@75)：
在terminal中运行
sudo dpkg-reconfigure xserver-xorg
然后根据提示设置，设定需要的分辨率。
之前还试过一种方法，先装5.10，然后再用alternative CD升级到6.06，可以成功。但有些机器上会出现xserver无法启动的问题。
今天又成功装了xubuntu6.06，这个发行版基于XFCE，号称轻量级速度很快，界面颜色不错，不过还是喜欢gnome。 November 05 this is one answer to many whyseveryone around me is extremely busy: running for deadlines, running for ACM contests, running for the course work.
manay times I could not leave my hands free, I ask myself one question in my deep heart:what am I seeking for? to aim at the real stuff, aim at the real impact we should create for the future, for the people around, and for the society. Now I am paving the way for the future. JUst as the MSN manager has said in the Wen. talk to ACM class, the Computer Science will continue a boom and it is we who are in responsibility of making great innovations, and revolutionizing the CS ideas as the MIT professor Victor Zue said.
What's more, I am deaply motivated by the impressive words of Xiao'ou:
|
|
|
