幸福的时空：阅读《复杂》

很久没有用中文写blog了，现在心中一种不可名状的兴奋让我感觉用中文来和大家分享更恰当一些。这就是我近两日阅读《复杂》的感觉。
晚上一位朋友来看我的位置看望我，提及《复杂》时他很惊讶，想必你一定和他一样。其实我读的这本书和我们计算机中平时讲的算法的复杂性几乎没有关系。这本书的全称是《复杂──走在秩序与混沌边缘》（Complexity: The Emerging Science at the Edge of Order and Chaos）。连接就在我这个blog左侧的阅读列表里面。
这是一本电子书，所以阅读的只能是晚上在研究院的业余时间，回到寝室就不能看了。前一阵子阅读这本书的感觉是在讲经济学中的混沌现象以及建立桑塔费研究所的令人兴奋的经过。现在我已经阅读完了前六章，还剩下三章。这本书的撰写方式我的感觉有点散，每出现一位传奇的人物都会转移话题到这个人上半天，好不容易才将话题转回原来，然后继续介绍。整本书我想更多的是在介绍复杂性非线性系统的创立背景，其中的桑塔费研究所成了必须介绍的机构。其中包含了很多精英科学家，因此自然也就要好好介绍他们的经历。当然整本书还是在围绕复杂性系统而展开的，经济学是其中涉及到的第一个实例。
我为之兴奋的是其中对那些牛人的传记介绍。这些牛人们有一个共同的特点。那就是，他们都在作他们为之兴奋的事情，完成他们的研究和学习完全是一种主动的事情。其中介绍了分子自动机，遗传算法，经济非稳定网络等有趣的东西。他们的工作完全都是自主的，而且经常出现自己的想法和别人的想法虽然不同领域但是却非常类似的情况。这让他们的研究很富有激情，而且经常出现灵感。书中崭露出的是研究人员的本性，他们的工作是富有创造性的，而且都能让人在兴奋状态下运作。
中午吃完饭和刚从VS转为正式员工的石磊聊了一会儿。他对我现在在做的工作非常有兴趣。出于保密我现在不能在我的blog中介绍我在这边的工作。这个事情非常具有挑战性。从阅读《复杂》的过程中我找到了很多灵感。我想继续发掘和深化我的想法，让我负责的模块有质的飞跃。
《复杂》真是本好书。可惜我阅读它都是断断续续的，等读完这一遍后我一定会再读几遍。其中的内容让我彷佛回到了小学和中学时对科学的那种由衷的崇拜，也让我认识到了从事科学研究的乐趣。我会继续的。

Swimming in four hours

Jianguo Du, Yi Chen and I had been swimming between 13:00 to 17:20. It's the longest period of my swimming history. I felt little tired. I believed I would get up late tomorrow morning.

I met an elder man again. I met him in my all four times swimming. He said he was 75 now and swam every afternoon. He could swim 1500 meters. We three all admired him.

Good health is the basis of all. I felt good at swimming.

Wonderful BBQ of MSRA

After long time preparation, the BBQ(barbeque) of MSRA had come this evening. We all gathered at the front of Sigma Building. Our goal was Central Garden, Friendship Hotel. When it was 19:00, all the MSRA interns and Miscrosoft Student Summer Camp attendees came here. Our evening started. There were nearly 500 persons.

At the beginning, I met the four MSClub students of our HIT. They were Zhongqi Yang, Yongqing Xiang, Chao Zhang, Junheng Zhang. They were the current leaders of MSClub of HIT. They had come in Beijing before four days and would return back to Harbin tomorrow morning. We were very happy to meet each other. We had many group photos. I wishes them nice trip.

There were many programs of this evening. Most were of singing. And some interesting games were hold. The most excited was that our Libra Poster for our Ping Pong Club had won the award of "the Best of Best". We five Libra guys were all excited. Harry was humorous. He trumped like us and said some jokes to us. My mentor Ming Zhou was very happy seating around our NLC VSs. At the end, there were two dancing programs. Ming Zhou went to the stage twice and played with us. He said that was very interesting.

During the activity, I chatted with some friends and had group photos with them. Xu Sun, who was a VS of NLC, said that he had seen me in SWCL last year in Beijing Language University. He was the student of Houfeng Wang, who was the leader in anaphora resolution I thought. We discussed some problem on anaphora resolution. I had known some research situation of Houfeng Wang's group on Anaphora resolution. I planed to visit Dr. Houfeng Wang. Now Xu Sun's work in BeiDa was on Abbreviation Resolution. It was very important for Anaphora resolution and Coreference Resolution. It seemed that their works on Anaphora resolution were in deep level. We discussed some related works.

It's a nice evening. Thanks to MSRA!

Wonderful Presentation of Study Group

This afternoon, Guihong Cao and Shenghua Bao gave us two wonderful presentations respectively.

Guihong's topic was on his SIGIR 05 paper: Integrating Word Relationships into Language Models. As it's his own paper, he presented it in very detail. The main idea of his paper was combining co-occurrence and WordNet for Information Retrieval. In his talk, I knew that Lemur had the ability of language modeling. And it had three smoothing techniques. I had heard that it could be used for Information Retrieval. However, I heard its language modeling ability firstly. It's interesting and useful. I believed that I should mater more tools for NLP research. There was another measure for evaluation of information retrieval. It was non-interpolated average precision. It was used in TREC Evaluation.

Shenghua gave us a talk on Comparative Study of Name Disambiguation Problem Using a Scalable Blocking-based Framework. This paper was on Name Disambiguation. I thought it was a subtype of coreference resolution and could be viewed as person name coreference resolution. The only feature used was co-author of papers. There was a heuristic idea that we could do coreference resolution by web information. It was newly and could do lots of topics research.

Thanks them for their so nice talks. Next time, it was my turn to give presentation. There was only 13 days left. I should prepare it day by day.

[Collection]Machine Learning大家(续)

发信人: Car (得之,我幸,不得,我命;如此而已), 信区: AI
标 题: Machine learning 大家（5): Michael Collins
发信站: 哈工大紫丁香 (Tue Aug 23 19:50:58 2005), 转信

Michael Collins (http://people.csail.mit.edu/mcollins/

自然语言处理(NLP)江湖的第一高人。出身Upenn，靠一身叫做Collins Parser的武功在江湖上展露头脚。当然除了资质好之外，其出身也帮了不少忙。早年一个叫做Mitchell P. Marcus的师傅传授了他一本葵花宝典-Penn Treebank。从此，Collins整日沉迷于此，终于练成盖世神功。

学成之后，Collins告别师傅开始闯荡江湖，投入了一个叫AT&T Labs Research的帮会，并有幸结识了Robert Schapire、Yoram Singer等众多高手。大家不要小瞧这个叫AT&T Labs Research的帮会，如果谁没有听过它的大名总该知道它的同父异母的兄弟Bell Labs吧: 言归正传，话说Collins在这里度过了3年快乐的时光。其间也奠定了其NLP江湖老大的地位。并且练就了Discriminative Reranking，Convolution Kernels，Discriminative Training Methods for Hidden Markov Models等多种绝技。然而，世事难料，怎奈由于帮会经营不善，这帮大牛又不会为帮会拼杀，终于被一脚踢开，大家如鸟兽散了。Schapire去了Princeton，Singer也回老家以色列了。Collins来到了MIT，成为了武林第一大帮的六袋长老，并教授一门叫做的Machine Learning Approaches for NLP (http://www.ai.mit.edu/courses/6.891-nlp/的功夫。虽然这一地位与其功力极不相符，但是这并没有打消Collins的积极性，通过其刻苦打拼，终于得到了一个
叫Sloan Research Fellow的头衔，并于今年7月，光荣的升任7袋Associate Professor。

在其下山短短7年时间内，Collins共获得了4次世界级武道大会冠军(EMNLP2002,
2004, UAI2004, 2005)。相信年轻的他，总有一天会一统丐帮，甚至整个江湖。


－－－－－－－－－－－－－－－－－－－－－－－－－－－－－－－－－－－－－
发信人: Car (得之,我幸,不得,我命;如此而已), 信区: AI
标 题: Machine learning 大家（6): Dan Roth
发信站: 哈工大紫丁香 (Wed Aug 24 18:52:49 2005), 转信

Dan Roth
http://l2r.cs.uiuc.edu/~danr/


统计NLP领域的众多学者后，我得出了一个惊人的结论，就是叫Daniel的牛人特别多:
大到MT领域成名已久的Prof. Dan Melamed，小到Stanford刚刚毕业的Dan Klein，中
间又有Dan jurafsky这种牛魔王，甚至Michael Collins的师弟Dan Bikel
(IBM Research)，ISI的Dan Marcu，获得过无数次TREC QA评比冠军的
Prof. Dan Moldovan (UTexas Dallas)，UC Berkeley毕业的Dan Gildea
(U Rochester)。但是，在众多的Dan中，我最崇拜的还是UIUC的Associate
Professor，其Cognitive Computation Group的头头Dan Roth。

这位老兄也是极其年轻的，Harvard博士毕业整十年，带领其团队撑起了UIUC Machine Learning以及NLP领域的一片灿烂天空。其领导开发的SNoW工具可谓是一把绝世好剑，基本达到了"又想马儿跑，又想马儿不吃草"的境界，在不损失分类精度的条件下，学习和预测速度空前。什么？你不知道SNoW？它和白雪公主有什么关系？看来我也得学学"超女"的粉丝们，来一个扫盲了: SNoW是Sparse Network of Winnows的简称，其中实现了Winnow算法，但是记住Sparse Network才是其重点，正是因为有了这块玄铁，SNoW之剑才会如此锋利。

近年来Roth也赶时髦，把触角伸向了Structured Data学习领域，但与其他人在学习的
时候就试图加入结构化信息(典型的如CRF)不同，Roth主张在预测的最后阶段加入约束
进行推理，这可以使的学习效率极大的提高，同时在某些应用上，还取得了更好的结果。
还有就是什么Kernel学习，估计他也是学生太多，安排不下了，所以只好开疆扩土。

Harvard出身的Roth，理论功底也极其深厚，好多涉及统计学习理论的工作就不是我这种学工科的人关心的了。

最后广播一条小道消息，Roth正在招Post-Dor，感兴趣的可以联系一下他，呵呵

Happy birthday to Jizhou and Shenghua

It was so coincident that Jizhou and Shenghua had the same birthday and in same group of MSRA.
We nine VSs had a nice dinner in Jiutouying. There were two related blogs about it: Jizhou and Xiaoyuan Cui

Good luck to them!

Know more clearly about my task

Before I had come here, I had not known my definite task. And during the past three months, I had done many dispersive works. In this month, I had done some research on dialogue. This afternoon, after our project weekly discussion, I had known my tasks and steps very clearly.

I believed it was a piece of good news for me. Then, I could do my research like my feeling on swimming. Let me just do the survey now!