|
|
2007/12/28
Human society is perhaps the most intricate system studied by scientists. The complex nature of human society, which mainly consists of limited degree of decomposability and emergence of structures and society level behaviours, is the major barrier prohibiting deeper understanding toward the underlying principles of societies by using completely analytical methods. To overcome this problem, methods based on Multi-Agent Simulation (MAS) have been developed.
An introduction to MAS based research in social sciences is given in this report, which was mainly dedicated for answering two questions that why MAS based methods are useful for social sciences research and how appropriate models can be developed. To give an answer to the first question, the notion of complex systems and the explanation that why human society is complex were first given. Then, the mathematical theory of System Dynamics has been introduced, which built a connection between analytical studies of complex systems and MAS based approaches and further highlighted that why simulation should be preferred. Finally, the second question was answered by a detailed review of MAS-based society models developed at Department of Computing (DoC) of Imperial College London, together with a paper survey of researches in Agent-based Computation Economics (ACE). Some inspirations that how these models can be combined to derive a more realistic simulation of the society have also been presented in this report. 2007/12/24 等洗衣机把衣服洗好就准备出门了，趁这会再随便写点。
一般到逢年过节的时候我就会想去年的这个时候我在干啥，前年在干啥，大前年在干啥……很无聊吧？其实过节本来就是挺无聊的事情，呵呵。
2006：在家宅了一天，虽然那段时间没去实验室，但还是一直在干活，当时在赶工增加了投影变换标定功能的规则编辑器。
2005：在浙大过的。正好那两个星期又回学校去弄出国申请的成绩证明材料了。好像跟CHZ一快去西湖边溜哒去了，中午在豪尚豪吃了顿牛排，挺爽。不过也可能是25号出去玩的。如果是这样的话，那24号肯定就是在曹东404的实验室里宅过去了。
2004：大四的这个时候在干什么呢？说实话还真记不清了，好像是在做专题实验的那个小玩意，也可能在寝室睡了一天，那年12月的后半个月够堕落的，谁知道呢。
2003：大三，电磁场与微波，CBY JJ的高频。看书、写作业、复习备考。和FD MM一起自修，表白被拒，哈，经典的平安夜：）
2002：大二，跟HDW还有协会的几个朋友去ZJG号称花了十多万装备的多媒体实验室看《终结者二》，音响效果超一流，记忆深刻。
2001：大一，大概在寝室看宏观经济学呢吧，那会真无聊，还写了很详细的读书笔记。
2000：高三，跟卷一起骑车回家，吃羊肉串和肉筋，调侃FZ和GMX。
以前的真记不清楚了。。。 其实和极光并没什么关系。刚才在新华网上看新闻，看到南极科考的图片报道中有好几张很漂亮的极光照片，于是就回想起了下面这个故事。
这个故事是在一集探索频道（Discovery Channel）的节目中看到的。故事说的是美国某个大学——好像还是南方某州的——里面一批专门研究极光的科学家。极光虽然经常有，但是光在地上看看可做不了研究。要想搞清楚发光的那个地方倒底发生了什么有意思的事儿，就得把科学仪器弄到产生极光的那快区域里去做实地测量。那可是距离地面70~100公里的高空，飞机和气球都飞不到那个高度。怎么办？只能靠探空火箭，于是这个组里有个博士生就花了两三年造了一个探空火箭。后来好不容易造好了，拿到加拿大那边去发射，打上去还没一会就掉下来了。火箭没飞到预定高度，什么数据都没收集到，仪器什么的还全摔烂了。那个博士生看着火箭掉下来，看他那表情都快疯了：自己博士这几年几乎就做了这么一个东西，结果什么作用都没发挥出来就摔烂了。
当时看到这我也快疯了…… 2007/12/19 从八岁的时候看了第一本阿西莫夫的作品《钢窟》，到上周末看完《机器人与银河帝国》，断断续续的看了十二年，我终于把阿西莫夫的基地系列、银河帝国系列和机器人系列全部看完了。电子版，硬拷贝；大陆翻译版，台译版，英文原版；短篇，中篇，长篇，一点一点把阿西莫夫为人类“规划”的未来两万年发展历程串起来，直到看完了《机器人与银河帝国》，所有的故事终于完美的扣到了一起，一部横跨数百代人、数千万个星球、辉宏壮丽的人类未来断代史跃然纸上。阿西莫夫果然不愧为有史以来最伟大的科幻作家之一，如此震撼的故事线，实在令人赞叹不已！
本来想精确的总结一下阿西莫夫的“未来简史”，但实在没时间去查具体年份，于是只好在这里凭记忆写一条粗略的时间线。因为是全凭记忆，其中难免有记错、记漏的部分，诸位如果看到哪里写的有问题，还请不吝赐教。最后还要说明一点，下文中时代的划分是我为了方便描述自己加上的，阿西莫夫的作品中并没有提到这一点。
1. 地球时代：1940s~2050s
阿西莫夫刚开始写《基地》的时候还是1940s，因此整个未来史的起点是从1940s开始的。换句话说，在阿西莫夫的“未来史”与真实历史的的分岔即开始与此。这也部分解释了为什么在阿西莫夫的作品中计算机一直是具有高度人工智能的庞然大物——这是那个时代的人们对于未来计算机的普遍幻想。
地球时代的开端为1948（或1949）年，标志性事件是一个正在散步的老头被一次实验意外制造出的神奇射线传送到了银河帝国中期的世界。这样一个开始似乎有些无厘头，但是因为这是我所记得阿西莫夫的所有相关作品中提到的最早的年份，因此就只好从这里算做故事的开端了。
地球时代终止时的标志性事件是美国机器人公司首席机器人心理学家苏珊·卡尔博士去世。
这个时代和我们目前所处的世界基本上差不太多。地球是人类唯一的定居星球，宇航技术与今天我们掌握的技术大致相当，人类尚未拥有跨星系航行的技术能力。在这一时期中，人类陆续在宇宙中建立了能源采集站，并在月球、太阳系内层行星和小行星带建立了矿站和科学考察站。地球上人口膨胀，资源被开采怠尽。世界上仍然存在不同的国家，但已逐渐融合成以洲划分的政治实体，联合国的权威日渐增大，有转变成为统一地球政府的趋势。对未来影响深远的机器人技术（正电子脑和机器人三大定律）诞生于这一时代。
这一时期发生的故事主要是机器人系列中的大部分短篇小说，比如《我，机器人》、《环舞》、《罗比》、《说谎话的机器人》、《他们那个时候多有趣》等。这些小说大多描写的是机器人受“机器人三大定律”约束的特殊“心理模式”所导致的一些有趣小故事，对于时代背景的宏观描写较少。
这一时期的主要机构包括：成立与1982年，专门生产机器人，几乎垄断市场的美国机器人公司。主要人物包括：美国机器人公司核心层人员，即机器人心理学家卡尔文博士、厂长兰宁博士、数学部主任以及一向运气不佳却又总能找到问题关键的两个现场测试工程师等人。
大事纪：
1948（或1949）年：整部“未来史”的开端，见前文。
1982年：美国机器人公司成立，机器人心理学家苏珊·卡尔文出生。
2003年：美国机器人公司创始人劳伦斯·罗泊逊发明正电子脑。这种全面超越传统电子电路技术的划时代发明使得让机器人拥有如同人一般的智能成为可能。从此机器人不再是机器，而更像是一种与人类拥有迥异思维模式的智慧生物。换句话说，人们可以制造机器人，但并不能完全把握机器人的行为，这也是“机器人心理学”这一学科出现的原因。为了从根本上保证机器人的有效性和安全性，所有的机器人都必须被写入一组最基本的行为准则，即“机器人三大定律”：1.机器人不得伤害人类，也不得在人类受到伤害时袖手旁观；2.在不违反第一条的前提下，机器人比如服从人类的一切命令；3.在不违反第一条和第二条的前提下，机器人必须保护自己。另外，出于某种原因（也可能是涉及这里的故事我没看到），法律禁止在地球上使用机器人。
2008年：苏珊·卡尔文自哥伦比亚大学毕业，获得哲学博士学位（研究方向为控制论），以机器人心理学家的身份加入美国机器人公司。
2020年左右：由于未知的制造流程意外，历史上第一台拥有读心术能力的机器人被制造出来。但它还没来得及对历史产生任何影响就因为伤害了老处女苏珊·卡尔文（时已将近40岁）的感情而被报废掉了。
2020s~2030s：在美国机器人公司的超级计算机的帮助下，跃迁驱动器被发明，并被不断加以改进。自此以后，由于拥有了几乎等同于瞬间移动的跨恒星系航行能力，向其它类地行星的殖民成为可能，人类得以进入第一次宇宙殖民时期。
2057年：苏珊·卡尔博士去世，享年75岁，终生未婚。 2007/12/18 一句话，太NB了。。。
引用
【转载】爱在西元前——信号与系统版本
爱在西元前－信号与系统版本
傅立叶研究物理提出了三角级数串
两百年前粗略的论断
催生傅立叶变换不朽的缠绵
我在z平面前
凝视系统函数的极点
祈祷你的收敛域回到我的单位圆
卷积 采样 滤波 重建
哪怕坐标已变换
在抽取插值间守护你不失真的容颜
拜读奥本海姆鸿篇
以线性系统之名许愿：
相爱 像二阶无阻尼振荡到永远
当信号只剩下离散的语言
DTFT就成了无尽的思念
我给你的爱藏在频谱间
在奈奎斯特率之下混叠难现
周期的重复被低通滤波还原
最初的响应依然隐约可见
我给你的爱藏在频谱间
在奈奎斯特率之下混叠难现
把快速傅立叶变换烧进芯片
蝶形流图美妙的弧线
引向谁心间
我感到很疲倦
离稳态并不远
害怕衰减的冲激串不能再重现
[原文由 doughtyR 发表于 飘渺水云间] 2007/12/17 The following is the first draft of my research statement prepared for PhD application. Since I have to submit my application ASAP, probably I'll not have enough time to wait for my supervisor's opinion. Therefore, I am pasting it here, please feel free to give comments, especially critiques, thank you!
Jie SHEN
1. Background
With the ever increasing demand of computerization in both industry and our everyday life, Multimodal Human Computer Interface (MHCI) has becoming a more and more emphasized research direction. Unlike most of other traditional fields in Computer Science which can be dated back to 1950s, the importance of MHCI research has not been raised before late 1980s, when Personal Computer began to enter majority’s daily life. Especially in the past decade, with the trend that usage of computer has been widely spread from the group of professionals to non-expert users, the limitation of traditional HCI has been clearly exposed. Firstly, it is already been approved by medical studies that using computer with traditional HCI (keyboard, mouse, screen, etc.) can irreversibly damage user’s health. Secondly, although with carefully designed GUI and other improvement techniques the usability and usefulness of traditional HCI can be increased to some extend, it is till generally unnatural and thus hard to be controlled by untrained users. For instance, users often claim that hundreds of menu items, buttons and dialog-boxes appeared in large software can easily get them frustrated. To overcome the limitation of traditional HCI, MHCI has been developed to offer a more natural, more effective, and more efficient way of controlling the computer.
2. Multimodal Human Computer Interface
Multimodal Human Computer Interface is currently not a strictly defined direction in Computer Science, thus a wide range of different research streams can be covered by this name. As it is known, the functions of HCI consists of a two-way communication, one way is to let the computer system understand user’s command and the opposite one is to let user perceive system’s output. For the later case, many approaches such as Holographic Display, Three Dimensional Projection and Augmented Reality have been suggested. Traditionally, these research streams are often being considered as separated directions, but it is worth to mention them here to help building a full-view of MHCI research. For the former case, a general definition of MHCI can be given as: the type of HCI which gathers information from user in more than one modality for the purpose of achieving more natural interaction between human and computer, in other words, more similar to human-human communication. The term modality refers to the channel carrying information from user to computer. From human-centered perspective, modalities correspond to human’s senses, including sight, hearing, touch, smell and taste. Traditional input methods which gather information from mouse, keyboard, etc. are also modalities, but a HCI exploits these channels only are often not being considered as MHCI since it is unnatural to user.
Most cases of human-human communication, especially when being conduced face-to-face, can be considered as an exchange of visual-audio signals via spoken language, gesture, facial expression, etc. Therefore, to achieve the goal of developing more natural HCI, Natural Language Processing and Computer Vision are solid foundation of MHCI research. The importance of Natural Language Processing is obvious, hence will not be further discussed here. By contrast, at the first glance, modality of vision may seem to be negligible. However, according to findings in psychology, 55% of information exchanged during face-to-face communication is carried by body language, visually. Hence, to understand user’s body language and thus user’s emotion by using facial expression recognition, gaze tracking, gesture recognition and other Computer Vision techniques is at the same importance of extracting user’s will from verbal context. In fact, Computer Vision based research is currently one of the dominating streams in the field MHCI.
3. Proposed Research Direction
My proposed research direction can be divided into two parts. The first part is to search for an appropriate approach to experiment oriented framework dedicated to MHCI research. The second part would be focusing on a specific Computer Vision branch closely related to MHCI applications, such as face detection / tracking or facial expression recognition.
Comparing to the popularity of the researches in specific Computer Vision algorithms, only few efforts have been made in the aspect of designing software frameworks for MHCI research. Absence of such a carefully designed framework may not be a severer problem at the beginning, but in the long-run, it will have significant negative effects on the study of MHCI. Firstly, it is obvious that without a proper designed framework, integrating separate algorithms into one applicable MHCI system would be hard to imagine. Secondly, research of high level algorithms would be cumbersome without proper mechanism of reusing low level modules, which can only be achieved by an incrementally architected system based on uniformly interfaced software framework. Finally, evaluation of algorithms would only be really accurate and useful if it is put into the context of a system as similar as possible to its desired application environment. Therefore, designing and implementing software framework for MHCI research would be a meaningful and promising direction.
To fulfill the goal presented above, several steps should be gone though progressively. A survey of frameworks which have already been used by researchers and developers in MHCI study or other similar fields should be done in the first place, to form a knowledge base of possible approaches. Then, in-depth study and analysis of several typical Computer Vision algorithms developed for MHCI applications would be necessary to achieve a design of the desired framework which is balanced between generality, efficiency and effectiveness. The actual designing and implementation should be conducted in the third phase, when varies Software Engineering methodology will be used. Evaluation and optimization of framework, as well as the algorithms’ performance within the framework, will also be done in the mean time.
Although a variety of algorithms have already been published by researcher to fulfill the need of MHCI applications, only few of them are effective and efficient enough for dealing with real-scene and real-time problems. Moreover, most of the successful algorithms are situated at low level. Therefore, to find better Computer Vision algorithms targeting to MHCI applications, especially at high level, is also a promising aspect, which will form the second part of my proposed research plan.
4. Relevant Experience
After graduation from Zhejiang University, I had been working for two years at National Laboratory of Pattern Recognition, CASIA as a research assistant, focusing on system development in the field of Intelligent Video Surveillance, which is an important application oriented branch of Computer Vision research. During this period, I have gathered much practical experience in both software development and Computer Vision, which may be crucial factors to the successful completion of my proposed PhD research plan. For details of my work experience and projects participated before, please refer to my Curriculum Vitae which is also provided with my application materials. 2007/12/16 今天着实睡了一回懒觉。其实早上睁开眼睛的时候并不晚，只不过又在床上赖了几个小时，所以一直到快11点才起床。既然是假期，那么懒就懒一点吧，用不着自责了。要不是因为躺着头痛，我真打算睡到12点然后直接起来做午饭，只是现在的自己实在没那个能耐了。起床，穿好衣服，开机，11:10。想想上午马上就要过去了，立刻就没了干正事的劲头，于是就又懒了一回，开始写Blog。
关于感悟
昨天跟LL同学讨论了一会对“感悟”的看法问题，发现自己当时并没有说清楚，所以这里再进一步解释一下。从“机器学习”的角度上说，所谓“感悟”其实也是一个学习过程，而且还是一种“热切学习”（Eager Learning），即：根据训练集（全体数据集合的一个有限子集）中的数据，基于一组假设，推导出全体数据集合应当服从的规律（Training Set & Inductive Bias |= Rules）。而“感悟”和一般学习过程的区别在于“感悟”基于的训练级更小，假设条件（Inductive Bias）更强，因此结论更不可靠。这就我不支持LL同学“感悟”泛滥的主要原因——它不可靠。
关于素质
中学的时候，我讨厌过的和讨厌（或鄙视）过我的老师都很多。尽管现在逐渐的也互相谅解了——或者说是我理解了“所应当被怨恨的是制度而不是人”这样一个道理，但是对于两位老师，我目前仍然无法有一点好的印象：语文徐ZC老师和英语汪Y老师。徐ZC老师的故事和本篇无关，不提了。我不喜欢汪Y老师的原因当然并非三言两语所能穷尽，但其中较主要（排在第二或第三）的一条是她似乎坚定的认为中国人素质远不及外国人，并如传教般坚持不懈在课堂上反复宣扬这一论断（此人乃英语老师，之前曾去过三个月美国）。
当时我对她的看法就不敢苟同，总觉有崇洋媚外之嫌，但那个时候我对所谓“国外”的概念不过是来自于媒体而已，所以也只能半信半疑的听着。如今我在资本主义阵营的老二国家前后也住了有两个多月，虽然断不敢说对国民素质的问题有了如何程度的认识——其一是我对社会学连半瓶醋都没有，其二是毕竟凭一己之力能接触到的社会范围完全不是统计抽样，但我想我也有资格发表一点我认为还算是比较公道的看法（勉强算是个案）了吧。
// 1点了，先干正事，回头再扯。 2007/12/15 Finally, all 11 weeks of the first term has become the past. At this particular check point, should I write something to form a conclusion or something? Probably it would be a good idea, but I have nothing to conclude so far, really. That is not because I have not exprienced anything worth to be written, but just the opposited. As gone through 11 weeks with full of shocks and acquisitions, I have too many things to be memorized that I hardly can decide which should be put in this article and which should be left out. Therefore, rather that to focus on the past, let me turn my face to the opposite direction along the time axis.
Vacation, indeed, a time to have a rest. So, the highest priority task would be: sleeping. haha, what a lazy guy I am. Well, as I am not that kind of unbelievables who can spend a whole day without leavng the bed, I should do something to kill my time when sun is above my head. But do what? Here are some great options to make vacation either cheerful or awful.
1. 30 pages ISO report. Without completing that, I'll certainly expect to meet my own end-of-the-world. So, no more comments need be put on thit issue.
2. Completion of the PhD application process. It may not be the end-of-the-world if I failed to finish this on time, but it won't be far from the senario since I'll be worrying about my near future in even nearer future. Therefore, this should be ranked second on my list. Moreover, the time pressure for this is also significant.
3. Do some reading about MMHCI. This is for the seond term's ISO, which may affect more than its mark. So I'd better be serious on that.
4. Revision of this term's courses. I really don't want my Eastern vaction to be overly excited. Hence, it would be a good idea to do some revision by now. Especially for the second half of Modal & Temporal Logic (syntactic analysis) and most parts of IDAPI, high intensity of readings would be necessary for that I have just caught very basic ideas but nothing else on them through the lectures.
5. At last, do I still have time to play? Well, certainly I will play regardless whether the answer to this question should goes to. The first stop will be the British Museum, today!
Alright, to reserve enough energy for today's great adventure though human civilizations, I'll get some sleep by now, so see you then.
2007/12/8 转眼之间已是第十周的周末，这个学期基本上就算是结束了。刚安顿下来似乎还是昨天的事情，现在却又要准备迎接“紧张而又充实”的寒假了。以前上大学的时候一到放假前总是很兴奋，这次却没有什么感觉。也许是因为没有多少考试吧。不过我想更主要的原因还是想到了这么一大笔的花销，实在兴奋不起来。
工作了两年，学术上进展并不大，最大的收获其实是知道了赚钱的不容易。没日没夜的干，家常便饭似的加班，到头来一个月也不过是赚点口粮钱。以前一个月的工资现在还不够两周的花销——这还不算学费。一想到这个，脑子里就开始不自觉的算帐。
一个星期的房租是XXX磅，合人民币XXXX+
一节课的平均费用是XX磅，合人民币XXX+
一个三明治的价格大概是X.X磅，合人民币XX+
一周地铁的花费是XX磅，合人民币XXX+
……
……
……
别的不说，XXXXX磅的学费，这就算是用去XXXX了。X万人民币啊，我工作了两年都没有赚到这个数。瞬间就没了。
当初在申请的时候其实真的没有想到太多关于钱的事情。母亲说家里会支持你至少你至少一年，只要你愿意读就去读吧。于是那个时候脑子里想的竟是学校的名次，专业的内容和自己的兴趣等等。我一直都知道来英国，来帝国理工读书要花多少钱，这笔帐我很早就估计过，只是我一直都在刻意忽视这个问题。第一次被迫正视这个数字是在申请签证的时候。表格里面问我预计读这个硕士要多少钱。XX万人民币，这还是保守估计。填到这里，我忽然就被震了一下。虽然我对这比花销早有估计，但是第一次认真的把这个数字写出来，笔尖还是有点颤。
四门课，一个小项目，若干次实验，几十页报告和作业。一个问题，一个答案。
值吗？
值。现在的情形不允许我怀疑这一点。我并不打算在这里重申什么要更加努力、好好读书之类的空话。有的人可以24小时都呆在图书馆里，但是我做不到，我不是这种人。
见鬼，我都写了点什么啊。算了，洗洗睡了，明天争取把KBO，RPO和Knuth Bendix过程搞明白。 2007/12/3 The good news is that I have finally found some interesting papers in this field. However, that bad news is that they are too many...
The reason why I had not gotten those papers in a week before is that I had searched using the wrong key words. Although we often use the term 'Multi-Agent Simulation' in the academia of Computing, the economists tend to use another name, called 'Agent-based Computational Economics', or ACE, instead.
There were three special issues focusing on this topic published in year 2001. More than 300 pages, or 20 papers, in total. It seems that I am going to spend a considerable amount of time to deal with my new discovery. Untill now, I have just finished the shortest issue, which contains 6 papers looking at the economy from a general perspective. Two of them which are relatively related to our approach implemented here looking at social science are presented below.
The first one by Epstein is focusing on social norms, which was also the field that Camille interested in. However, rather than simulating and analyzing the self-enforcing feature of social norms which is the main theme dominating the research in this field (incluing Camille's), Epstein's paper was focused on another feature of social norms, which is suggested by Epstein as 'once entrenched, we conform without thinking about it'. Then, Epstein further claims that his model and experiments indicate that individual thought is inversely related to the strength of norm.
In his research, the model is rather simple (much simpler than Camille's). The world is simulated as a ring - the globe in one dimensional case. Each agent has two parameters, ones is his norm, and another one is his 'thoughtfulness'. The norm of an agent is a boolean variable which simply denotes which norm the agents takes as granted. The 'thoughfulness' is represented as an integer. In his the model, 'thoughfulness' means how far an agent would see when deciding which social norm to obtain. Therefore, it is also called radius. For example, if this parameter is set to 3, then the agent would obtain his own behaviour norm by refering to 6 (3 to the left and 3 to the right) nearest agents' norm. The updating rule for the two internal parameters are also trivial. An agent's social norm is decided solely by taking the majority-accepted one in his neighbourhood defined by his radius. The radius is updated so that the agent would behave as a 'lazy statist'. In other words, the agent would always tend decrease the radius if the smaller sample set indicates the same relative frequency of different norms.
Epstein has tested his model on several situations. One is the 'locked in' pattern, where all agenets are set to the same norm initially. As the result, this norm would donimate the entire history of the society, and every agent would mindlessly follow the norm after a small number of iteration at the beginning. (To be continued...)
/* Not finished yet, but I got to sleep by now, so see you then */
Things will coming in this article:
A Complete introduction to Epstein's work on social norms.
A Complete introduction to Wilhite's work on trade network.
My impression so far about the current research streams in the field of ACE.
(Today's photo, Wan-Shou-Si Temple, Beijing, shot on 09/06/2007)
|
工具 
%3a+2007%e5%b9%b412%e6%9c%88&referrer=){kind=small}
