好久没有更新我的blog了，其实有很多东西想记录下来。
今天刚刚参加完MSRA和清华大学联合举办的internet service workshop multimedia session, 我讲了今年上半年的一个工作，transductive multi-labeling learning for video concept detection, 之后我们有大概一个小时左右的free discussion, 认识了不少内地高校的老师，西安交大的，复旦的，天津大学的，还有武大的，这些老师有做multimeida的，有做computer vision的，做learning的有，不过很少。一个比较可惜的事，就是似乎没怎么见到的清华的，当时也邀请了我在清华的导师，可是他正好有事没有能够参加，甚为遗憾。然后晚上我们team 5个人，和来自香港，台湾，新加坡的三个professor还有另外2个内地professor，在海底捞腐败了一下。
忘了今天比较有意思的事情。winston见到说，guojun qi都要毕业了，你还没有毕业啊，然后碰到西安交大的老师也问我是不是intern。看来我浑身散发的还是学生气。
这段时间一直比较忙，星期二刚刚忙完了WWW的paper，然后又开始今年的CVPR，可惜的是，到现在为止我也只是写了几行的abstract，不过还好核心算法部分应该基本ready了，只是怎么组织起来的问题，我想虽然还有2个星期不到的时间，但是应该是能够顺利完成的，给自己设个deadline吧，下周二完成introduction，然后周末完成所有的东西，比较好的是，下周似乎没有什么琐碎的事，会耽搁paper writing。
特别想抽时间仔细谈谈工作的感受。这里就概括说几句吧。总的说来，我更加喜欢现在的research topics，更喜欢这里的氛围，... 不过啊，还有个不同的地方，运动的机会少了，以前在学校一个星期可以打三次的羽毛球，现在一周只能挤出一次了。说到羽毛球，又想到了HKUST那些球友们了，和球友们打球愉快的时光，甚为想念，...
Friday, November 7, 2008
Monday, September 22, 2008
Wednesday, September 3, 2008
Thursday, July 31, 2008
Sunday, July 27, 2008
Wednesday, July 9, 2008
Sunday, July 6, 2008
随笔
周六晚上打羽毛球,有两个体会:
1) 体力下降。开始的时候跟一个球友打单打，异常激烈。我们打到26:25的时候，我发现自己体力支撑不住了，眼前感觉有点像以前跑长跑时最后快要晕的样子，只能在领先的情况下，自己放弃了。
2) 生活单调。一个球友说，他快要结婚了，不过自己不太想，因为结婚了晚上就需要天天回去，要按时回去，夜生活就缺少了。我发现我似乎从来没有这样的问题，因为我们经常加班，晚回去晚是正常的，平时很少有夜生活的问题。
1) 体力下降。开始的时候跟一个球友打单打，异常激烈。我们打到26:25的时候，我发现自己体力支撑不住了，眼前感觉有点像以前跑长跑时最后快要晕的样子，只能在领先的情况下，自己放弃了。
2) 生活单调。一个球友说，他快要结婚了，不过自己不太想，因为结婚了晚上就需要天天回去，要按时回去，夜生活就缺少了。我发现我似乎从来没有这样的问题，因为我们经常加班，晚回去晚是正常的，平时很少有夜生活的问题。
Friday, June 27, 2008
Surprise On Google
When I added last post about the rejection of my ACM MM08 submission. After 4 miniutes, I googled "ACM MM 08", I was very amazing to find my last post was indexed by Google.Just now, I found it very interesing and regoogled it and printscreen the result page, and pasted in on the right.
My submission got outside the ACM MM 08
Very disappointed! .......
The meta review:
"
to accept - if room.
From the three reviews, it appears that the article is not very appealing but the is no serious argument to reject it as it includes all required elements for a fair acceptation (except maybe the section on results).
"
The meta review:
"
to accept - if room.
From the three reviews, it appears that the article is not very appealing but the is no serious argument to reject it as it includes all required elements for a fair acceptation (except maybe the section on results).
"
Wednesday, June 25, 2008
Tuesday, June 24, 2008
Sunday, June 22, 2008
ZZ: 我的博士后老板
发信人: Thermophile (Jay), 信区: Biology
标 题: 我的博士后老板
发信站: BBS 未名空间站 (Thu Jun 19 21:00:28 2008)
BSO
********************************
我的博士后老板
K是我的博士后老板，成为他的弟子是一个偶然。八年前，我来到 美国的一所二 流大
学，在生物领域念博士。博士生活过的很颓废，导师经常生病在家，基本上没有研究可
以做，组里的学生除了我之外，都转走了。四年后，导师决定退休， 让我匆匆毕业。
我的博士论文就是把那些先前跑掉的学生留下的数据综合一下， 倒也有厚厚的一本。
因为博士期间没有发表过任何论文，也没有好的研究背景，所以毕业后拖了整整一个学
期，也找不到工作。当我已准备收拾行李回中国时，偶然在网上搜到了K的主页，他是
一个国家实验室生物工程部门的负责人。我给他发了一封email，他居然回了信，而且
给了我一个10个月的博士后合同，薪水只有三万一年，但我已经非常满意了。
我在K指导下工作了三年半，仿佛从头把phD补做了一遍。因为以前没有做过很多bench
实验，许多技能都很生疏。博士后工作第一个月就出了两个大事故，做PCR时，把一瓶
gel拧紧了盖子用微波炉加热，发生了爆炸，差点伤了人。接着，发酵实验中，反应器
的冷凝水接口没有密封好，在半夜时，发生了泄漏，大水淹了半个实验室。此外，还不
小心弄坏了一台离心机和一个气相质普的柱子，害的同事怨声连天，自己也很惶惶不安
，觉得很快会被解雇掉。一天，K 果然把我单独叫到办公室，但出乎意料的是他告诉我
说，博士后和学生一样，是需要时间学习的，所以特地通知了两个有经验的技术员，
让我多向他们请教。两个月后，K又把我送到能源部的PNNL 和Oak Ridge 学习细胞培
养和质普分离技术。在K 组 的头一年，几乎每天都在学习各种新技术。虽然没有做出
任何有意义的结果，K 还是给我续了一年约。 到了博士后的第二年，当研究工作开始
变得熟练起来时。K还让我接了几个不同领域的side projects， 涉及到医疗，生物能
源和纳米技术的研究。每一个领域对我来说都是陌生的，但是从每个项目中都能学到很
多新的东西，极大的开阔了研究视野。
K 常说，做学生时，最重要的是hardwork, 做PI 时，最重要的是network。K在学校，
公司和national labs都有很多合作者，基本上我所有的课题 都是和别的研究单位一起
做的。K 很看重人品（personality），因为personality 是 建立network 的关键。
这也是他任聘人的主要标准。曾经实验室来过一个MIT毕业的博士后，有很多文章，非
常能干，却被K解雇了，因为他不能和同事和睦相处。K说，他不期望组里的每个人都“
be productive”, 但希望每个人都“enjoy their team”。 所以组里的气氛不错，每
年年底都会评选几个乐于助人的先进。我的第一篇paper是关于一种新的生物反应器研
制。这个课题是和K的一个公司合作的.我的工作就是在两天的细胞生长时间内，帮公司
每隔三小时取样测量在不同反应条件下的一些参数，基本上是个体力活。博士后期间，
我非常幸运的和许多不同的研究小组合作过， 甚至去过包括Sandia这样敏感的单位。
而这些合作使很复杂的课题变得相对简单，也容易出成果。这些network 还是K 留给我
的一笔无形的财富。
K讲过，做生物的不需要很聪明，但对科研的心态一定要好， 也就是要有passion。我
phD毕业时，已快30岁了，生活和事业都是空白的，觉得自己是个loser。可是到了K组
里，发现K也没有结过婚，每周7天，早上7点就来办公室上班，有时晚上10点还在工作
。科学事业就是他生活的全部。K每天路过我的办公桌时，都会对我打招呼“hello,
Jay”。看到K 的身影，我总会有一种莫名的感动和激励，因为K那慈祥的眼神总使我想
起过逝的父亲，也许觉得现在所有的努力是对我过去的颓废的一种救赎。三年多来，除
了星期天会花两个小时去附近的教堂里听牧师讲道外，我基本上都在实验室工作, 从来
没有觉到厌倦，因为K也是这样生活的。K的研究大局观很强，他为我选择的课题和指导
的思路都很简洁，也有新意，所以我的研究很少走弯路，渐渐的也有了一种成就感。
有时候觉得, 人生实际上是不需要完美的，只要能够一直去做有意义的事业，生命就会
很充实。
三年为限，K从不希望他的博士后 做的太久而耽误了前程。于是，他给了三个月的
时间让我全心找工作。因为我的口语很差，K 帮我把面试要讲的seminar talk ，chalk
talk和可能要被问及的问题答案都写下来，又修改了一遍，再 让我把它们背诵下来。
我知道K 给我写了很好的推荐信，因为我的工作申请和面试过程都非常顺利。离开K组
的时候， K还送给了我一些菌种和他用过的教学材料。
K的组除我之外几乎没有中国人，至今不知道为何K在四年前会让我去他的实验室工作，
也许这就是缘份吧。 但是这段经历改变了我一生的道路。我希望以后也会像K一样，能
给我的学生一个机会，一个经历，和一个光明的未来。
标 题: 我的博士后老板
发信站: BBS 未名空间站 (Thu Jun 19 21:00:28 2008)
BSO
********************************
我的博士后老板
K是我的博士后老板，成为他的弟子是一个偶然。八年前，我来到 美国的一所二 流大
学，在生物领域念博士。博士生活过的很颓废，导师经常生病在家，基本上没有研究可
以做，组里的学生除了我之外，都转走了。四年后，导师决定退休， 让我匆匆毕业。
我的博士论文就是把那些先前跑掉的学生留下的数据综合一下， 倒也有厚厚的一本。
因为博士期间没有发表过任何论文，也没有好的研究背景，所以毕业后拖了整整一个学
期，也找不到工作。当我已准备收拾行李回中国时，偶然在网上搜到了K的主页，他是
一个国家实验室生物工程部门的负责人。我给他发了一封email，他居然回了信，而且
给了我一个10个月的博士后合同，薪水只有三万一年，但我已经非常满意了。
我在K指导下工作了三年半，仿佛从头把phD补做了一遍。因为以前没有做过很多bench
实验，许多技能都很生疏。博士后工作第一个月就出了两个大事故，做PCR时，把一瓶
gel拧紧了盖子用微波炉加热，发生了爆炸，差点伤了人。接着，发酵实验中，反应器
的冷凝水接口没有密封好，在半夜时，发生了泄漏，大水淹了半个实验室。此外，还不
小心弄坏了一台离心机和一个气相质普的柱子，害的同事怨声连天，自己也很惶惶不安
，觉得很快会被解雇掉。一天，K 果然把我单独叫到办公室，但出乎意料的是他告诉我
说，博士后和学生一样，是需要时间学习的，所以特地通知了两个有经验的技术员，
让我多向他们请教。两个月后，K又把我送到能源部的PNNL 和Oak Ridge 学习细胞培
养和质普分离技术。在K 组 的头一年，几乎每天都在学习各种新技术。虽然没有做出
任何有意义的结果，K 还是给我续了一年约。 到了博士后的第二年，当研究工作开始
变得熟练起来时。K还让我接了几个不同领域的side projects， 涉及到医疗，生物能
源和纳米技术的研究。每一个领域对我来说都是陌生的，但是从每个项目中都能学到很
多新的东西，极大的开阔了研究视野。
K 常说，做学生时，最重要的是hardwork, 做PI 时，最重要的是network。K在学校，
公司和national labs都有很多合作者，基本上我所有的课题 都是和别的研究单位一起
做的。K 很看重人品（personality），因为personality 是 建立network 的关键。
这也是他任聘人的主要标准。曾经实验室来过一个MIT毕业的博士后，有很多文章，非
常能干，却被K解雇了，因为他不能和同事和睦相处。K说，他不期望组里的每个人都“
be productive”, 但希望每个人都“enjoy their team”。 所以组里的气氛不错，每
年年底都会评选几个乐于助人的先进。我的第一篇paper是关于一种新的生物反应器研
制。这个课题是和K的一个公司合作的.我的工作就是在两天的细胞生长时间内，帮公司
每隔三小时取样测量在不同反应条件下的一些参数，基本上是个体力活。博士后期间，
我非常幸运的和许多不同的研究小组合作过， 甚至去过包括Sandia这样敏感的单位。
而这些合作使很复杂的课题变得相对简单，也容易出成果。这些network 还是K 留给我
的一笔无形的财富。
K讲过，做生物的不需要很聪明，但对科研的心态一定要好， 也就是要有passion。我
phD毕业时，已快30岁了，生活和事业都是空白的，觉得自己是个loser。可是到了K组
里，发现K也没有结过婚，每周7天，早上7点就来办公室上班，有时晚上10点还在工作
。科学事业就是他生活的全部。K每天路过我的办公桌时，都会对我打招呼“hello,
Jay”。看到K 的身影，我总会有一种莫名的感动和激励，因为K那慈祥的眼神总使我想
起过逝的父亲，也许觉得现在所有的努力是对我过去的颓废的一种救赎。三年多来，除
了星期天会花两个小时去附近的教堂里听牧师讲道外，我基本上都在实验室工作, 从来
没有觉到厌倦，因为K也是这样生活的。K的研究大局观很强，他为我选择的课题和指导
的思路都很简洁，也有新意，所以我的研究很少走弯路，渐渐的也有了一种成就感。
有时候觉得, 人生实际上是不需要完美的，只要能够一直去做有意义的事业，生命就会
很充实。
三年为限，K从不希望他的博士后 做的太久而耽误了前程。于是，他给了三个月的
时间让我全心找工作。因为我的口语很差，K 帮我把面试要讲的seminar talk ，chalk
talk和可能要被问及的问题答案都写下来，又修改了一遍，再 让我把它们背诵下来。
我知道K 给我写了很好的推荐信，因为我的工作申请和面试过程都非常顺利。离开K组
的时候， K还送给了我一些菌种和他用过的教学材料。
K的组除我之外几乎没有中国人，至今不知道为何K在四年前会让我去他的实验室工作，
也许这就是缘份吧。 但是这段经历改变了我一生的道路。我希望以后也会像K一样，能
给我的学生一个机会，一个经历，和一个光明的未来。
Tuesday, June 17, 2008
Tex: How to make a LaTeX poster
How to make a LaTeX poster
1. Making the poster file (almost from [1, 2, 3])
1) Download template latex source file UFposter.tex, blank figure ps file blank.ps, and UF logo ps file LetterShape.eps
2) Texify it o UFposter.div, and Divps the div file to Fposter.ps
3) In GSview, File/Convert, select device “jpg” and resolution “300”, and save it as a jpg file.
2. Special hints
1) If we look at the ps file, we can observe that the heights of different columns may be different, which affects the visual quality. One solution is to change the code segment by adding the red text.
\newenvironment{kasten }{%
\begin{lrbox}{\dummybox}%
\begin{minipage}{0.96\linewidth} }%
{\end{minipage}%
\end{lrbox}%
\raisebox{-depth}{\psshadowbox[framesep=1em]{\usebox{\dummybox}}}\\[0.5em]}
\newenvironment{kasten}{%
\begin{lrbox}{\dummybox}%
\begin{minipage}[t][22.6cm]{0.96\linewidth} }%
{\end{minipage}%
\end{lrbox}%
\raisebox{-depth}{\psshadowbox[framesep=1em]{\usebox{\dummybox}}}\\[0.5em]}
2) For convenience, we usually donot set the physical size in the tex file and set the proper size so that it is easy to visually check in the pdf viewer, for instance
%\def\breite{390mm} % Special Format.
%\def\hoehe{319.2mm} % Scaled by 2.82 this gives 110cm x 90cm
%\def\anzspalten{4}
%\def\breite{600mm} % Special Format.
%\def\hoehe{300mm} % Scaled by 2.82 this gives 110cm x 90cm
%\def\anzspalten{4}
\def\breite{420mm} % A3 LANDSCAPE
\def\hoehe{297mm}
\def\anzspalten{4}
% \def\breite{297mm} % A3 PORTRAIT
% \def\hoehe{420mm}
% \def\anzspalten{3}
Hence, it is necessary to resize the ps poster file. The script poster_resize offers a solution. However, the script can only be run in Unix. For windows, we just psresize this ps file to the size that we want.
An example command:
psresize -w118cm -h168cm -W29.7cm -H42cm in.ps out.ps.
Comments: -w -h specify the physical size, -W –H correspond to the size specified in the tex file. Strangely, the width and height specification seems exchanged, maybe for the reason of landscape or portrait view.
Wednesday, June 4, 2008
Thursday, May 22, 2008
提高工作效率
最近感觉到时间利用效率不够高.下面摘抄了一篇关于效率的文章,希望对自己有帮助.
现代人的生活节奏越来越快，压力也越来越大。经常会听到白领人士抱怨一个星期有三到四天的时间在加班，没有时间锻炼身体，身体经常处在一种透支的状态；也有人抱怨，虽然现在的职位已经到了中层管理层，但是没有安全感，因为知识的更新速度太快。其实大家都感觉到时间是个瓶颈，每天列了一大堆的计划，晚上回顾的时候，却发现忙的都是一些琐碎的事情，重要的事情却没有干，这都是因为缺乏时间管理的技能，不能很好的运筹时间。时间是世界上最充分的资源，每个人都拥有24小时的一天，然而时间又是世界上最稀缺的资源。这就要求我们必须懂得充分地利用每一分每一秒的时间。大多数成功者都有一个共同特点，他们都是管理时间的高手，而失败的管理者则无一例外地都不善于管理时间，这是因为，管理时间是有技巧的。
下面介绍几种高效管理时间的方法。
一、优先计划管理(1) 把每天要做的事列一份清单(2) 确定优先顺序，从最重要的事情做起(3) 每天都这么做大家可以看到这三条建议的关键点在于根据事情的重要程度决定优先顺序。为什么要根据事情重要程度来决定优先顺序？举一个例子：有一个非常大的桶具，桶旁边放了一些大石块、小石块、水和沙子，怎么做能最大限度的把这些东西都放进桶里？正确的顺序是先放大石块再放小石块，再放沙子，最后放水，如果按照相反的顺序，先把水倒进去，再放沙子，再放小石块，最后放大石块，这个桶就装不下这么多的东西。其实，人的精力就像这个桶的容量一样是有限的。大石块就相当于那些非常重要的事情，那些小石块、沙子和水其实就相当于琐碎的小事。如果先去处理那些琐碎的小事，到最后大的事情反而会被忽略到一边。所以要按照事情的重要程度来确定优先顺序，这样能够节省你的精力和时间，重点专注于你要做的事情。
二、目标管理很多人都说，定目标就一定有收获，那么制定目标究竟有哪些好处呢？ (1) 通过制定目标，可以节省不少时间。(2) 目标就像一个筛子，可以筛选出要做的事情，过滤掉无关的事情。如果这件事情和目标有关系就去做，和目标没关系就不必做，这样生活会井然有序，并且有目的。(3) 目标就像风向标一样，它能够使你沿着既定方向前进，并唤起你最大的潜能，推动你实现人生价值。成功人士都是那些目标远大并且能够付诸实施的人。有调查显示，100％的人都认为定目标是有好处的，但只有２０％的人清楚自己具体的目标，能清楚地描述出他想要做的事情。这２０％的人中只有３％的人把目标写下来，变成书面的东西。这个调查结果表明，这３％的人无论是从收入还是成就上都要比那97％的人高。大家都知道定目标好，但是在制定目标的时候还是存在很多的误区。那么，到底应该如何制定目标呢？制定目标要遵循一定的原则，否则目标就会很空。制定目标的原则有以下几条：(1)目标必须自己定。只有你自己参与，原动力才会很大。因为只有你自己清楚，你想要做什么，要达到什么样的标准。(2)目标必须切合实际。如果定得太低，没有挑战性，不能激发潜能，就没有任何意义；如果定得太高，往往会中途流产。有句话说得好，目标只要踮点脚尖就能够得着，就是合理的；如果不用踮脚尖就够到，那就没有挑战性；跳起来都够不到，那就不是目标。 (3)目标一定是书面的。要向３％的人学习，目标书面化可以理清思路，时常提醒要实现的事情，就不容易遗忘。(4)目标必须是可衡量的。只有可衡量的目标，才能知道这个目标到底实现了没有，如果不是可衡量的目标，就很难成为行动的指南。(5)目标必须是有时限的。如果定了一个毫无时间限制的目标，那就不称其为目标了。只有有时限的目标，才可以拟制相应的行动计划。(6)目标是平衡和谐的。你制定的目标可能有很多个，包括生活上的、工作上的，这些目标相互间不能冲突，否则就会使你无所适从。
现代人的生活节奏越来越快，压力也越来越大。经常会听到白领人士抱怨一个星期有三到四天的时间在加班，没有时间锻炼身体，身体经常处在一种透支的状态；也有人抱怨，虽然现在的职位已经到了中层管理层，但是没有安全感，因为知识的更新速度太快。其实大家都感觉到时间是个瓶颈，每天列了一大堆的计划，晚上回顾的时候，却发现忙的都是一些琐碎的事情，重要的事情却没有干，这都是因为缺乏时间管理的技能，不能很好的运筹时间。时间是世界上最充分的资源，每个人都拥有24小时的一天，然而时间又是世界上最稀缺的资源。这就要求我们必须懂得充分地利用每一分每一秒的时间。大多数成功者都有一个共同特点，他们都是管理时间的高手，而失败的管理者则无一例外地都不善于管理时间，这是因为，管理时间是有技巧的。
下面介绍几种高效管理时间的方法。
一、优先计划管理(1) 把每天要做的事列一份清单(2) 确定优先顺序，从最重要的事情做起(3) 每天都这么做大家可以看到这三条建议的关键点在于根据事情的重要程度决定优先顺序。为什么要根据事情重要程度来决定优先顺序？举一个例子：有一个非常大的桶具，桶旁边放了一些大石块、小石块、水和沙子，怎么做能最大限度的把这些东西都放进桶里？正确的顺序是先放大石块再放小石块，再放沙子，最后放水，如果按照相反的顺序，先把水倒进去，再放沙子，再放小石块，最后放大石块，这个桶就装不下这么多的东西。其实，人的精力就像这个桶的容量一样是有限的。大石块就相当于那些非常重要的事情，那些小石块、沙子和水其实就相当于琐碎的小事。如果先去处理那些琐碎的小事，到最后大的事情反而会被忽略到一边。所以要按照事情的重要程度来确定优先顺序，这样能够节省你的精力和时间，重点专注于你要做的事情。
二、目标管理很多人都说，定目标就一定有收获，那么制定目标究竟有哪些好处呢？ (1) 通过制定目标，可以节省不少时间。(2) 目标就像一个筛子，可以筛选出要做的事情，过滤掉无关的事情。如果这件事情和目标有关系就去做，和目标没关系就不必做，这样生活会井然有序，并且有目的。(3) 目标就像风向标一样，它能够使你沿着既定方向前进，并唤起你最大的潜能，推动你实现人生价值。成功人士都是那些目标远大并且能够付诸实施的人。有调查显示，100％的人都认为定目标是有好处的，但只有２０％的人清楚自己具体的目标，能清楚地描述出他想要做的事情。这２０％的人中只有３％的人把目标写下来，变成书面的东西。这个调查结果表明，这３％的人无论是从收入还是成就上都要比那97％的人高。大家都知道定目标好，但是在制定目标的时候还是存在很多的误区。那么，到底应该如何制定目标呢？制定目标要遵循一定的原则，否则目标就会很空。制定目标的原则有以下几条：(1)目标必须自己定。只有你自己参与，原动力才会很大。因为只有你自己清楚，你想要做什么，要达到什么样的标准。(2)目标必须切合实际。如果定得太低，没有挑战性，不能激发潜能，就没有任何意义；如果定得太高，往往会中途流产。有句话说得好，目标只要踮点脚尖就能够得着，就是合理的；如果不用踮脚尖就够到，那就没有挑战性；跳起来都够不到，那就不是目标。 (3)目标一定是书面的。要向３％的人学习，目标书面化可以理清思路，时常提醒要实现的事情，就不容易遗忘。(4)目标必须是可衡量的。只有可衡量的目标，才能知道这个目标到底实现了没有，如果不是可衡量的目标，就很难成为行动的指南。(5)目标必须是有时限的。如果定了一个毫无时间限制的目标，那就不称其为目标了。只有有时限的目标，才可以拟制相应的行动计划。(6)目标是平衡和谐的。你制定的目标可能有很多个，包括生活上的、工作上的，这些目标相互间不能冲突，否则就会使你无所适从。
Wednesday, May 21, 2008
Machine Learning Links
1. Machine Learning Lunch seminar at Carnegie Mellon
http://www.cs.cmu.edu/~learning/
2. Video Lectures
http://videolectures.net/Top/Computer_Science/Machine_Learning/
3. scholarpedia
http://www.scholarpedia.org/
4. Reinforcement Learning
http://www.cs.ualberta.ca/~sutton/book/ebook/the-book.html
5. Matrix Cookbook
http://matrixcookbook.com/
6. Hunch
http://hunch.net/
http://www.cs.cmu.edu/~learning/
2. Video Lectures
http://videolectures.net/Top/Computer_Science/Machine_Learning/
3. scholarpedia
http://www.scholarpedia.org/
4. Reinforcement Learning
http://www.cs.ualberta.ca/~sutton/book/ebook/the-book.html
5. Matrix Cookbook
http://matrixcookbook.com/
6. Hunch
http://hunch.net/
Wednesday, May 14, 2008
Friday, May 9, 2008
Google VisualRank
Google VisualRank
Google researchers at a web conference in Beijing announced they work on some kind of PageRank specifically aimed at images. Called VisualRank, the technology was so far only applied to a smaller test set of images, as apparently applying it to all images Google indexed would be too computing-intensive (even arguably the world’s largest super-computer can’t do everything imaginable yet). According to the New York Times yesterday, visual rank is an algorithm “for blending image-recognition software methods with techniques for weighting and ranking images that look most similar,” and in Google’s internal scoring tests it achieved far higher quality results.
If I understand the gist of the research paper [PDF] right, then it seems the core of Google’s VisualRank algo consists of not only looking at textual cues in regards to images, but also image content itself. After identifying the most authoritative set of picture candidates for a given query, Google then improves the ranking of images found to be sharing the most visual characteristics with the group at large, by creating a similarity network (which also would understand e.g. imagery shown from different perspectives, to a certain extent). Center node images or those images containing large resolution versions would then determined to be the most relevant. In 1000 sample queries – taken from the top Google Product search queries – 762 VisualRank results were tested to be more relevant than Google’s old approach, with 202 equal quality results and only 70 results that were worse.
(On a side-note, I wonder what motivates Google to publicize this information, as it could tip of their competition? Are they only being nice, potentially attracting more good researchers, or is there more to it?)
As far as I can tell, the paper does not yet indicate that Google is any further in specific image recognition, e.g. figuring out that the image found on the web contains, say, a vase of flowers without looking at textual descriptions in the vicinity of the embedded pic. 4 years ago, Google co-founder Sergey Brin said, “I don’t think that in the near future we’re going to have a service that takes a picture, and the computer decides, oh, that’s an elephant, so we search for an elephant. That seems funny to us. We should be able to do it.” Google does have face recognition features for Google Images, though; it’s found in the advanced image search dialog and works very well, and you can also use it on your own site, if you have one with indexed images, by searching Google Images for site:yourdomain.com with the face search activated.
Google researchers at a web conference in Beijing announced they work on some kind of PageRank specifically aimed at images. Called VisualRank, the technology was so far only applied to a smaller test set of images, as apparently applying it to all images Google indexed would be too computing-intensive (even arguably the world’s largest super-computer can’t do everything imaginable yet). According to the New York Times yesterday, visual rank is an algorithm “for blending image-recognition software methods with techniques for weighting and ranking images that look most similar,” and in Google’s internal scoring tests it achieved far higher quality results.
If I understand the gist of the research paper [PDF] right, then it seems the core of Google’s VisualRank algo consists of not only looking at textual cues in regards to images, but also image content itself. After identifying the most authoritative set of picture candidates for a given query, Google then improves the ranking of images found to be sharing the most visual characteristics with the group at large, by creating a similarity network (which also would understand e.g. imagery shown from different perspectives, to a certain extent). Center node images or those images containing large resolution versions would then determined to be the most relevant. In 1000 sample queries – taken from the top Google Product search queries – 762 VisualRank results were tested to be more relevant than Google’s old approach, with 202 equal quality results and only 70 results that were worse.
(On a side-note, I wonder what motivates Google to publicize this information, as it could tip of their competition? Are they only being nice, potentially attracting more good researchers, or is there more to it?)
As far as I can tell, the paper does not yet indicate that Google is any further in specific image recognition, e.g. figuring out that the image found on the web contains, say, a vase of flowers without looking at textual descriptions in the vicinity of the embedded pic. 4 years ago, Google co-founder Sergey Brin said, “I don’t think that in the near future we’re going to have a service that takes a picture, and the computer decides, oh, that’s an elephant, so we search for an elephant. That seems funny to us. We should be able to do it.” Google does have face recognition features for Google Images, though; it’s found in the advanced image search dialog and works very well, and you can also use it on your own site, if you have one with indexed images, by searching Google Images for site:yourdomain.com with the face search activated.
Image-search startup Riya calls Google's plans "largely impossible"
Image-search startup Riya calls Google's plans "largely impossible"
Google-backed researchers Shumeet Baluja (pictured) and Yushi Jing presented the Mountain View company's latest image search and recognition efforts to an audience in Beijing, China on Thursday. VisualRank attempts to do for images what PageRank has done for typical Web pages — rank them in search results according to "authority," which will presumably increase the relevance of results. Problem is, their limited success came at a cost Google is typically loathe to pay: 150 units of homo sapiens who helped sort and rank the images by hand. Munjal Shah, CEO of image-search startup Riya, remarked to the Times: "I think what they're trying to accomplish is largely impossible." Funny, because large-scale, advanced image recognition is what Marissa Mayer says will solve Street View's privacy conundrum.
Google-backed researchers Shumeet Baluja (pictured) and Yushi Jing presented the Mountain View company's latest image search and recognition efforts to an audience in Beijing, China on Thursday. VisualRank attempts to do for images what PageRank has done for typical Web pages — rank them in search results according to "authority," which will presumably increase the relevance of results. Problem is, their limited success came at a cost Google is typically loathe to pay: 150 units of homo sapiens who helped sort and rank the images by hand. Munjal Shah, CEO of image-search startup Riya, remarked to the Times: "I think what they're trying to accomplish is largely impossible." Funny, because large-scale, advanced image recognition is what Marissa Mayer says will solve Street View's privacy conundrum.
Thursday, May 8, 2008
Monday, April 28, 2008
Tuesday, April 22, 2008
Matlab figure tips
http://www.mathworks.com/access/helpdesk/help/techdoc/index.html?/access/helpdesk/help/techdoc/creating_plots/bqrw9tj.html&http://www.mathworks.com/access/helpdesk/help/techdoc/helptoc.html
http://www.mathworks.com/access/helpdesk/help/techdoc/index.html?/access/helpdesk/help/techdoc/creating_plots/f10-19972.html&http://www.mathworks.com/access/helpdesk/help/techdoc/ref/bar.html
http://www.mathworks.com/access/helpdesk/help/techdoc/index.html?/access/helpdesk/help/techdoc/ref/bar.html&http://www.mathworks.com/access/helpdesk/help/techdoc/creating_plots/f9-53405.html
http://www.mathworks.com/matlabcentral/files/3486/xticklabel_rotate.m
http://www.mathworks.com/access/helpdesk/help/techdoc/index.html?/access/helpdesk/help/techdoc/creating_plots/f10-19972.html&http://www.mathworks.com/access/helpdesk/help/techdoc/ref/bar.html
http://www.mathworks.com/access/helpdesk/help/techdoc/index.html?/access/helpdesk/help/techdoc/ref/bar.html&http://www.mathworks.com/access/helpdesk/help/techdoc/creating_plots/f9-53405.html
http://www.mathworks.com/matlabcentral/files/3486/xticklabel_rotate.m
Monday, April 21, 2008
The increasing velocity of the paper number
“The increasing velocity of the paper number is higher than the speed of light, but there is nothing to worry about for there is no violation of any physical law, because these papers carry no information.”
by Chandrasekhar
by Chandrasekhar
Sunday, April 20, 2008
Good articles (I): 我是职业人
结果证明价值
责任造就人品
能力解决问题
重视日常细节
耐心对待客户
遵守公司规范
团队利益为重
精确时间观念
沟通解决一切
行为始终如一
http://www.zhiyeren.cn/book/index_2.asp
第一章 永远以结果证明自己的价值， 职场没有苦劳，只有功劳
http://www.zhiyeren.cn/discussGroup/DG_TopicView.asp?tid=666
第二章 永远以责任心证明自己的人品，忠诚于事业，坚守承诺
http://www.zhiyeren.cn/discussGroup/DG_TopicView.asp?tid=667
第 3 章 永远把解决问题的技能作为核心能力，而不是把知识丰富作为核心能力
责任造就人品
能力解决问题
重视日常细节
耐心对待客户
遵守公司规范
团队利益为重
精确时间观念
沟通解决一切
行为始终如一
http://www.zhiyeren.cn/book/index_2.asp
第一章 永远以结果证明自己的价值， 职场没有苦劳，只有功劳
http://www.zhiyeren.cn/discussGroup/DG_TopicView.asp?tid=666
第二章 永远以责任心证明自己的人品，忠诚于事业，坚守承诺
http://www.zhiyeren.cn/discussGroup/DG_TopicView.asp?tid=667
第 3 章 永远把解决问题的技能作为核心能力，而不是把知识丰富作为核心能力
Thursday, April 17, 2008
PracTex Journal
Avoid eqnarray: http://www.tug.org/pracjourn/2006-4/madsen/
The PracTexJournal: http://www.tug.org/pracjourn/archive.html
The PracTexJournal: http://www.tug.org/pracjourn/archive.html
Authorship Practice Guideline
I thought every researcher and student ought to know what the right authorship practice is, but from my observations and conversations with people in MSRA, this does not seem to be the case. Since some of you asked me to write down a guideline, here it is.
To decide if it is worth your time to read this, try to see if you fit any one of these symptoms:
You ever put someone on your paper or patent even though that someone has very little or no technical contribution.
Your name ever appeared on a paper or patent even though you have very little or no technical contribution.
All your published papers so far have more than 2 authors.
If so, you might be hurting yourself. Reasons and remedies are provided below.
The Rules
1. authors contain only people with sufficient technical contributions.
By technical contribution, I mean efforts directly attributed to the paper, including idea, algorithm, coding, demo, experiments, and paper writing. A university professor who merely sent his/her student(s) to work on a project does not qualify for authorship. Anyone (e.g. a manager or a colleague) who does nothing or merely comments on a paper (or patent) draft does not quality for authorship.
2. authors should be ordered according to their relative contributions.
It is not always easy to judge the relative contribution of the collaborators, but my rule of thumb is idea > algorithm/writing > coding/demo > labor (e.g. data collection and experiments). If you have difficulty figuring this out, a good mental exercise is to ask how replaceable that particular author is; the more replaceable, the less value. (This is how supply & demand work in a market economy.)
FAQ
Why is it bad to put a student programmer the first author?
(1) It is not fair, as I stated above. (2) It is harmful to the student's growth. If the student can be a first author by merely coding, what would motivate him/her to go further? (3) It is bad for you (researchers), as in the end you will have a bunch of students or junior employees who know nothing beyond coding, and you will have to do all the other stuff yourself. It is usually not fun.
Why is it bad to put a "guest" professor (i.e. a professor that does nothing to a paper beyond sending students) as an author?
(1) It is not fair, as I stated above. (2) It is harmful to that professor, as he/she would be less motivated to do real research. Plus, putting your name on a paper you know little can be very dangerous to your reputation; remember the American professor who put his name on one of the papers fabricated by the infamous Korean stem cell researcher? (3) We, MSRA, are not doing our job to promote the research capability of China and Asia. Quote from Harry Shum: "it is better to teach someone how to fish rather than giving him free fish".
But that professor will no longer send students if I don't put his/her name on the paper.
I understand this is the Achilles' Heel of MSRA. Ultimately, I don't see this as a major issue as many of you will eventually become (at least) an adjunct professor and will be able to have your own students. In the meanwhile, if you need to bend the rules, go ahead. (I have been and I am still bending the rules myself even at the time of writing this article. We all need to compromise with the reality.) But just keep in mind that this is NOT right and you should strive to fix it eventually.
Why is it bad to put a manager or a colleague (who does not do enough work) as an author?
(1) It is not fair, as I stated above. (2) It can be harmful for the manager's reputation if he/she doesn't know enough about the paper (e.g. a paper containing fabricated results). It can also harm the manager's reputation in another way, as people will know he/she is "rubber-stamping". In the end, people will respect that manager as a powerful political figure, but not an intellectual researcher. (3) It is harmful to you. If you put managers on all your papers, people will not recognize it as your contribution (unless you are already famous to begin with).
On a related note: managers are paid for their "management" work via salary and compensation, not through (your) gratuitous authorship.
But I need other people's help to publish a paper
This implies that your research skill set is not yet complete. Fortunately, most of these skills are learnable. Even for paper writing, you don't have to travel abroad to learn it; I know at least one guy who picked up writing entirely by repeated practice while busting his ass inside Sigma building.
Related Issues
patent invalidation
Abusing authorship for a research paper is never good, but at least it won't be lethal. But doing that for a patent is. I have some recent conversations with several patent lawyers, and my understanding is that patent authors need to be exactly those who have actually made technical contributions (as itemized in the claims); otherwise, the patent can be invalidated in a court. A guy in MSR Redmond even told me that he believed many MSRA patents are subject to invalidation due to the authorship malpractice by managers.
too many authors on a paper
Try to answer this quiz: 10 authors on 10 papers; which option is better for their reputation? (1) 10 authors appear on each one of the 10 papers, (2) each person single-authors one paper. Mathematically they seem to be the same but they have drastically different effects on your reputation. I am not saying that massive collaboration is not good, but research is a very personal endeavor. If you want to get recognition, you will need papers that are associated with you, not your colleagues, manager(s) or institution. Whenever I hear people mention "a paper coming from a bunch of Chinese guys in MSRA", I know it is bad news for *each one* of these guys.
Implementation
Students and junior researchers should start as 2nd or 3rd authors, and they can become a first author only until the day they can finish the paper all alone, with only high level guidance from the adviser. (Sounds harsh? By definition, this is how a Ph.D. thesis requires.)
Senior researchers should start to embark on more personal projects and try not to have everybody collaborate on every project (or worse, gratuitous authorship). I believe the best practice is "one mentor, one student". My experience is that even for projects that have more than 2 people, usually one student and I have done most of the work.
For "guest" professors and managers, it is an art to get things right. A "stop-loss" strategy is to restrict the abuse to a small and fixed number of people that you know well, and cultivate those who are willing to make changes (if needed, feel free to share my article with them). The best strategy, in my opinion, is to become a professor yourself and have your own students.
Finale
In trying to make this article as concise as possible, I might have omitted important information (that I took for granted). Please let me know if you have any questions. I will be happy to give an informal talk if enough of you express interests.
To put my money where my mouth is, let's play a game. For the first MSRA student or assistant/associate researcher who single-authors a SIGGRAPH paper, your travel expense to the conference is on me. (Sorry, only SIGGRAPH can make my eyes blink. Volunteers are needed for other conferences.) The offer is valid throughout my lifetime even after I leave MSRA. Up to the challenge?
To decide if it is worth your time to read this, try to see if you fit any one of these symptoms:
You ever put someone on your paper or patent even though that someone has very little or no technical contribution.
Your name ever appeared on a paper or patent even though you have very little or no technical contribution.
All your published papers so far have more than 2 authors.
If so, you might be hurting yourself. Reasons and remedies are provided below.
The Rules
1. authors contain only people with sufficient technical contributions.
By technical contribution, I mean efforts directly attributed to the paper, including idea, algorithm, coding, demo, experiments, and paper writing. A university professor who merely sent his/her student(s) to work on a project does not qualify for authorship. Anyone (e.g. a manager or a colleague) who does nothing or merely comments on a paper (or patent) draft does not quality for authorship.
2. authors should be ordered according to their relative contributions.
It is not always easy to judge the relative contribution of the collaborators, but my rule of thumb is idea > algorithm/writing > coding/demo > labor (e.g. data collection and experiments). If you have difficulty figuring this out, a good mental exercise is to ask how replaceable that particular author is; the more replaceable, the less value. (This is how supply & demand work in a market economy.)
FAQ
Why is it bad to put a student programmer the first author?
(1) It is not fair, as I stated above. (2) It is harmful to the student's growth. If the student can be a first author by merely coding, what would motivate him/her to go further? (3) It is bad for you (researchers), as in the end you will have a bunch of students or junior employees who know nothing beyond coding, and you will have to do all the other stuff yourself. It is usually not fun.
Why is it bad to put a "guest" professor (i.e. a professor that does nothing to a paper beyond sending students) as an author?
(1) It is not fair, as I stated above. (2) It is harmful to that professor, as he/she would be less motivated to do real research. Plus, putting your name on a paper you know little can be very dangerous to your reputation; remember the American professor who put his name on one of the papers fabricated by the infamous Korean stem cell researcher? (3) We, MSRA, are not doing our job to promote the research capability of China and Asia. Quote from Harry Shum: "it is better to teach someone how to fish rather than giving him free fish".
But that professor will no longer send students if I don't put his/her name on the paper.
I understand this is the Achilles' Heel of MSRA. Ultimately, I don't see this as a major issue as many of you will eventually become (at least) an adjunct professor and will be able to have your own students. In the meanwhile, if you need to bend the rules, go ahead. (I have been and I am still bending the rules myself even at the time of writing this article. We all need to compromise with the reality.) But just keep in mind that this is NOT right and you should strive to fix it eventually.
Why is it bad to put a manager or a colleague (who does not do enough work) as an author?
(1) It is not fair, as I stated above. (2) It can be harmful for the manager's reputation if he/she doesn't know enough about the paper (e.g. a paper containing fabricated results). It can also harm the manager's reputation in another way, as people will know he/she is "rubber-stamping". In the end, people will respect that manager as a powerful political figure, but not an intellectual researcher. (3) It is harmful to you. If you put managers on all your papers, people will not recognize it as your contribution (unless you are already famous to begin with).
On a related note: managers are paid for their "management" work via salary and compensation, not through (your) gratuitous authorship.
But I need other people's help to publish a paper
This implies that your research skill set is not yet complete. Fortunately, most of these skills are learnable. Even for paper writing, you don't have to travel abroad to learn it; I know at least one guy who picked up writing entirely by repeated practice while busting his ass inside Sigma building.
Related Issues
patent invalidation
Abusing authorship for a research paper is never good, but at least it won't be lethal. But doing that for a patent is. I have some recent conversations with several patent lawyers, and my understanding is that patent authors need to be exactly those who have actually made technical contributions (as itemized in the claims); otherwise, the patent can be invalidated in a court. A guy in MSR Redmond even told me that he believed many MSRA patents are subject to invalidation due to the authorship malpractice by managers.
too many authors on a paper
Try to answer this quiz: 10 authors on 10 papers; which option is better for their reputation? (1) 10 authors appear on each one of the 10 papers, (2) each person single-authors one paper. Mathematically they seem to be the same but they have drastically different effects on your reputation. I am not saying that massive collaboration is not good, but research is a very personal endeavor. If you want to get recognition, you will need papers that are associated with you, not your colleagues, manager(s) or institution. Whenever I hear people mention "a paper coming from a bunch of Chinese guys in MSRA", I know it is bad news for *each one* of these guys.
Implementation
Students and junior researchers should start as 2nd or 3rd authors, and they can become a first author only until the day they can finish the paper all alone, with only high level guidance from the adviser. (Sounds harsh? By definition, this is how a Ph.D. thesis requires.)
Senior researchers should start to embark on more personal projects and try not to have everybody collaborate on every project (or worse, gratuitous authorship). I believe the best practice is "one mentor, one student". My experience is that even for projects that have more than 2 people, usually one student and I have done most of the work.
For "guest" professors and managers, it is an art to get things right. A "stop-loss" strategy is to restrict the abuse to a small and fixed number of people that you know well, and cultivate those who are willing to make changes (if needed, feel free to share my article with them). The best strategy, in my opinion, is to become a professor yourself and have your own students.
Finale
In trying to make this article as concise as possible, I might have omitted important information (that I took for granted). Please let me know if you have any questions. I will be happy to give an informal talk if enough of you express interests.
To put my money where my mouth is, let's play a game. For the first MSRA student or assistant/associate researcher who single-authors a SIGGRAPH paper, your travel expense to the conference is on me. (Sorry, only SIGGRAPH can make my eyes blink. Volunteers are needed for other conferences.) The offer is valid throughout my lifetime even after I leave MSRA. Up to the challenge?
Wednesday, April 16, 2008
Search Links
1. AltaVista
http://www.altavista.com/
2. Retrievr
http://labs.systemone.at/retrievr/
3. Riya
http://www.riya.com/
4. Blinkx
http://www.blinkx.com/
5. citeseerx
http://citeseerx.ist.psu.edu
http://www.altavista.com/
2. Retrievr
http://labs.systemone.at/retrievr/
3. Riya
http://www.riya.com/
4. Blinkx
http://www.blinkx.com/
5. citeseerx
http://citeseerx.ist.psu.edu
Tuesday, April 15, 2008
ACM MM08: deadline extended
As a tradition, the ACM MM08 deadline is extended.
Good luck to all ACM MM08 guys.
Good luck to all ACM MM08 guys.
Friday, April 11, 2008
Wednesday, April 9, 2008
Yahoo's photo-sharing site, Flickr, expands into video with new service
Tuesday, April 08, 2008
Yahoo's photo-sharing site, Flickr, expands into video with new service
By Associated Press
http://www.technologyreview.com/Wire/20544/?nlid=997
SAN FRANCISCO (AP) -- Yahoo Inc. will begin showing homemade videos on its online photo-sharing site, Flickr, in a long-anticipated move that may be too late to lure most people away from the Internet's dominant video channel, Google Inc.'s YouTube.
Flickr's video technology, to debut late Tuesday, represents the latest example of Yahoo trying to catch up to Google in a crucial battleground.
Yahoo's inability to keep pace with Google in the lucrative online search market caused its profits and stock price to sag during the past two years, which in turn triggered an unsolicited takeover attempt by Microsoft Corp. for more than $40 billion (euro25.5 billion).
While trying to fend off Microsoft, Yahoo has continued to develop and introduce services that the Sunnyvale-based company hopes will help revive its earnings growth.
Unlike Internet search, online video has not blossomed into a big moneymaker yet. But it is expected to turn into a marketing magnet as advertisers shift more of their spending from television in pursuit of consumers who are watching more entertainment and news online.
Yahoo already operates one of the Web's largest video platforms, but most of its content is provided by media outlets and other outside professionals.
Flickr's new technology is aimed at amateurs and hobbyists looking for a better way to share short video clips with family and friends.
Only Flickr's ''pro'' members -- those who pay for a $24.95 annual subscription -- will be allowed to transfer video clips of up to 90 seconds to the site, but anyone will be able to watch them. A privacy setting will allow videographers to limit access to the clips on Flickr if they want.
The video service will be offered in English and seven other languages: French, German, Italian, Korean, Portuguese, Spanish and traditional Chinese.
Flickr believes its service will offer a more personal touch than the many other Web sites that feature video, and that will help distinguish it. Flickr managers also expect to appeal to people looking to keep their video and pictures on the same site.
''What we are doing is going to meet a huge unmet need in the market,'' predicted Kakul Srivastava, Flickr's general manager. ''Most people aren't showing their personal videos at all right now.''
A trio of friends -- Chad Hurley, Steve Chen and Jawed Karim -- created YouTube partly because they couldn't find a spot on the Internet to share their personal videos. Shortly before YouTube played its first clip in the spring of 2005, the husband-and-wife team of Caterina Fake and Stewart Butterfield sold Flickr to Yahoo for a reported $35 million (euro22.3 million).
While Flickr continued to focus on photos, YouTube's eclectic mix of professional clips -- often illegally posted -- and videos of kids goofing off turned the site into a cultural phenomenon.
Since Google bought YouTube for $1.76 billion in late 2006, the video-sharing site has become even more popular despite increased competition from major media companies like NBC Universal and News Corp.
In February, 70 million people in the United States watched 2.9 billion video clips on YouTube, according to the research firm Nielsen Online. News Corp.'s Fox Interactive Video showed about 406 million clips to 21 million people in the United States to rank a distant second. Yahoo was in third place with 245 million clips shown to 21 million people in the United States.
Flickr has built a fiercely loyal following in its own right, a factor that should bolster its expansion into video.
The site attracted a worldwide audience of 42 million in February, up 53 percent from the same time last year, according to comScore Media Metrix. Facebook.com, an online hangout popular among younger Web surfers, had the only larger photo-sharing service with a worldwide audience of 65 million, comScore said.
Yahoo's photo-sharing site, Flickr, expands into video with new service
By Associated Press
http://www.technologyreview.com/Wire/20544/?nlid=997
SAN FRANCISCO (AP) -- Yahoo Inc. will begin showing homemade videos on its online photo-sharing site, Flickr, in a long-anticipated move that may be too late to lure most people away from the Internet's dominant video channel, Google Inc.'s YouTube.
Flickr's video technology, to debut late Tuesday, represents the latest example of Yahoo trying to catch up to Google in a crucial battleground.
Yahoo's inability to keep pace with Google in the lucrative online search market caused its profits and stock price to sag during the past two years, which in turn triggered an unsolicited takeover attempt by Microsoft Corp. for more than $40 billion (euro25.5 billion).
While trying to fend off Microsoft, Yahoo has continued to develop and introduce services that the Sunnyvale-based company hopes will help revive its earnings growth.
Unlike Internet search, online video has not blossomed into a big moneymaker yet. But it is expected to turn into a marketing magnet as advertisers shift more of their spending from television in pursuit of consumers who are watching more entertainment and news online.
Yahoo already operates one of the Web's largest video platforms, but most of its content is provided by media outlets and other outside professionals.
Flickr's new technology is aimed at amateurs and hobbyists looking for a better way to share short video clips with family and friends.
Only Flickr's ''pro'' members -- those who pay for a $24.95 annual subscription -- will be allowed to transfer video clips of up to 90 seconds to the site, but anyone will be able to watch them. A privacy setting will allow videographers to limit access to the clips on Flickr if they want.
The video service will be offered in English and seven other languages: French, German, Italian, Korean, Portuguese, Spanish and traditional Chinese.
Flickr believes its service will offer a more personal touch than the many other Web sites that feature video, and that will help distinguish it. Flickr managers also expect to appeal to people looking to keep their video and pictures on the same site.
''What we are doing is going to meet a huge unmet need in the market,'' predicted Kakul Srivastava, Flickr's general manager. ''Most people aren't showing their personal videos at all right now.''
A trio of friends -- Chad Hurley, Steve Chen and Jawed Karim -- created YouTube partly because they couldn't find a spot on the Internet to share their personal videos. Shortly before YouTube played its first clip in the spring of 2005, the husband-and-wife team of Caterina Fake and Stewart Butterfield sold Flickr to Yahoo for a reported $35 million (euro22.3 million).
While Flickr continued to focus on photos, YouTube's eclectic mix of professional clips -- often illegally posted -- and videos of kids goofing off turned the site into a cultural phenomenon.
Since Google bought YouTube for $1.76 billion in late 2006, the video-sharing site has become even more popular despite increased competition from major media companies like NBC Universal and News Corp.
In February, 70 million people in the United States watched 2.9 billion video clips on YouTube, according to the research firm Nielsen Online. News Corp.'s Fox Interactive Video showed about 406 million clips to 21 million people in the United States to rank a distant second. Yahoo was in third place with 245 million clips shown to 21 million people in the United States.
Flickr has built a fiercely loyal following in its own right, a factor that should bolster its expansion into video.
The site attracted a worldwide audience of 42 million in February, up 53 percent from the same time last year, according to comScore Media Metrix. Facebook.com, an online hangout popular among younger Web surfers, had the only larger photo-sharing service with a worldwide audience of 65 million, comScore said.
An interesting string pattern analysis problem
Given a binary string S 01...010 with N bits, and a string comparison operator Op(s1, s2) with outputing only 1 bit value. Suppose the binary string S have K "1".
Please give an algorithm to decide the positions of all the bits valued 1 and analyze its time compleixty in terms of operation. Can we even analyze the overall time complexity with the time complexity of operation.
Please give an algorithm to decide the positions of all the bits valued 1 and analyze its time compleixty in terms of operation. Can we even analyze the overall time complexity with the time complexity of operation.
Two passes in siggraph08 camera-ready paper
Just now, I heard a message about preparing siggraph08 camera-ready paper.
Interestingly, there are two passes:
Interestingly, there are two passes:
- Before April 21st, the author should submit a "draft" version for a camera-ready paper. Then some feedbacks are presented to the authors.
- Before April 30th, the author should submit the "final" version as the camera-ready paper.
Tuesday, April 8, 2008
Welcome to my academic blog
In my blog, I will collect a set of resources about my interseted research areas, including machine learning, computer vision and graphics, and multimedia.
Subscribe to:
Posts (Atom)
