主题：【原创】围绕脑科学而发生的若干玄想 -- 鸿乾

1.

"Stanford在sentiment analysis上貌似做出了突破性的工作，用RNN训练的模型精度达到了85%，号称搞定了情感分析（nailed sentiment analysis），新闻报道见 http://t.cn/zRAKhcy 而且他们将要公布源代码 http://t.cn/z8xuao1 看来深度学习在自然语言处理领域的春天要到来了！ "

"现在一般用矩阵当算子来做向量的composition吧？Richard Socher,Chris Manning发了一批利用word vector composition做NLP的文章，包括上次看到的情感分析那一篇，没记错的话。以前还看过向量加、乘、外积的composition方法，看来路子不少。词向量有了，组合方法也有了。好顶赞"

http://gigaom.com/2013/10/03/stanford-researchers-to-open-source-model-they-say-has-nailed-sentiment-analysis/

天津大学可视化小组：近日，斯担福大学的研究生Richard Socher和Andrew Ng（Google深度学习项目工程师），和一位语言学及人工智能领域的专家Chris Manning，共同开发了一个深度学习算法Neural Analysis of Sentiment，缩写为NaSent。NaSent算法从人脑中得到灵感，改善了分析方法，准确率达到了85%。

NaSent情感分析项目由斯坦福大学的研究生 Richard Socher 发起，合作者包括人工智能研究员 Chris Manning，以及 Google 深度学习项目的工程师之一 Andrew Ng。 准确率是 85%。建立了一个实时演示的网站。如果 NaSent 的判断错误，人们可以对其判断进行重新标记-http://t.cn/z86HYSl

“人们很好心地教它新东西，告诉它正确与错误，”Socher 说，“给出实时演示的好处是，人们试图去破坏它。他们在把它推向极限，给予我们新的训练数据。这会帮助我们的算法模型。”

the "code" part of the model is all open to public as well

2.

the core of stanford's model

“You’ll never care about translating a single word to another single word,” he said. ”We’re actually able to put whole sentences and longer phrases into vector spaces without ignoring the order of the words.”

3.

thanks for your "quote", see some similarities between Stanford model & human brain?

科学家发现大脑海马体绘图抄写“一心二用”

来源：中国科技网-科技日报作者：常丽君2013年11月30日　01:20

[导读] 最近,一个由美国宾夕法尼亚大学和德国弗莱堡大学神经科学家组成的研究小组,通过让志愿者玩一种“送货”游戏后发现,神经元能编码空间信息,给一段特定记忆标注“地理标记”,并能在这段记忆被回想前立即激活。

科技日报讯（记者常丽君）据物理学家组织网11月29日(北京时间)报道，最近，一个由美国宾夕法尼亚大学和德国弗莱堡大学神经科学家组成的研究小组，通过让志愿者玩一种“送货”游戏后发现，神经元能编码空间信息，给一段特定记忆标注“地理标记”，并能在这段记忆被回想前立即激活。相关论文发表在最近出版的《科学》杂志上。

该研究显示了空间信息是怎样被纳入记忆的，以及人们在回想一次经历时，为何还会迅速想起曾在同一地点发生过的其他事情。宾夕法尼亚文理学院心理学教授迈克尔·卡哈纳说：“有种观点认为，人类记忆系统会用记忆形成时的地点、时间来给记忆做‘标记’，而回忆涉及到还原这些‘标记’。我们的发现为此提供了首个直接的神经证据。”

参与实验的志愿者是颅内植入电极的癫痫病人，玩游戏时，电极能捕获他们的脑电活动。研究人员先让他们学习城市布局和各商店位置。游戏开始时，他们只被告知下一站要到哪里，而不知道送的是什么货，到达目的地后才会知道，然后再到下一站。这样送了13次货以后，屏幕变成空白，让他们尽可能多地回忆他们刚才到哪里送了哪些货。这样，研究人员就把神经活动与空间记忆形成（商店位置）、情节回忆（递送的货物清单）联系在一起。

在辨别方向过程中，海马体及其附近脑区的神经元通常会表征志愿者在城市里的虚拟位置，就像一种大脑的GPS设备。这些所谓的“地点细胞”可能是神经元编码一个抽象识别物的最典型样本。利用志愿者辨别方向时的脑电记录，研究人员开发出一份与城市布局相符的神经地图，并参考志愿者在回想送货情节记忆时的空间记忆后发现，在志愿者想到一个某地所送的货名后，与该地点有关的地点神经元立即被激活。

“我们正在观察地点细胞激活在回忆过程中的作用。我们发现，自发回忆一段记忆会激活它的神经‘地标’，这表明海马体的空间记忆与情节记忆密切相关，或许反映了它可能是一个公共功能结构。”卡哈纳说。

总编辑圈点

大脑的海马体就像是计算机的内存，早在上世纪初，就开始有科学家认识到海马体对于某些记忆学习的基本作用。而现在神经学研究认为，海马体其实演绎着双面的角色：一位是“绘图员”，负责跟踪空间记忆的位置信息；另一位是“抄写员”，负责记录情节记忆的事件。本次的实验则提供了进一步证据，证明海马体这两个功能是交织在一起的

本篇文章来源于: 中国科技网 www.wokeji.com

原文链接：http://www.stdaily.com/qyts/1_qykj/201311/t20131130_594669.shtml

4.

vector spaces etc, math (including physics) is becoming "social science" now

I have posted quite bit on 微分几何 etc, about the importance of 公理化 methodology of modern 物理 & math, & its applications in AI etc;

human brain's algos, 公理化 methodology of modern 物理 & math, logic of market place & other social organizations & human processes, etc, they will be all eventually "programmed" out, and thrown in your face, making it looks like 1+1=2, then it becomes 公理 on internet

5.

the immediate app is in the area of human communications:

how much time we have wasted in understanding each other first, then communicating to each other?

an example:

"I" don't care you are 公知 or 五毛, just tell me your position, so I can trade with you;

of course, very often, on a particular subject, we don't even know if we are 公知 or 五毛, or a mixed;

6.

a "公知五毛" AI app

1)

will help you read through "BS" in front of you, such as my this post, so you can filter through "BS" faster , knowing what it really talks about, then form your own comments, as 公知, 五毛, or both, etc ;

2)

"I" will then understand you a lot of better & faster with help of my similar "AI" in the cloud as well

3)

then we can do business together, after an short, "hot abusive insulting", brief "思想交锋" & wrap it up in an hour, instead of 1 day, for example, then move on, “求同存异”, move on to money-making business transactions, or say good by to a just gossip type of conversations of political entertainment, etc

7.

quantum jump in society's 劳动生产率

for almost of us, we live and work as some kind of "social scientists", whatever our profession may be, on the job, or as a consumer, voter, or as 老百姓心中有杆秤 in china.

8.

society's 劳动生产率 bottle neck is not really in physics/math or even in "manufacturing" area, it is in those "social science" area, where the "思想交锋 AI"breakthrough will be happening, leveraging on modern physics/math powered AI modeling & technology

bottom line: if we can't wipe out physically each other via the conventional "hot" war, then we have to “求同存异”, do business together, then we may just do the “求同存异” part faster, more efficiently & effectively;

now, as I have posted zillions times in the past, where will china be in this comming AI economy of

global capitalism's "upgraded phase"?

9.

do Chinese top science institutions such as CAS and top political institutions such as Mr. X and his red generations II buddy group even have a clue about all these potentially "non-linear" jump/changes of human capitalism, & their potential earth shaking implications?

for an example, how should TG prepare for an possibly "AI 茉莉花 revolution"? omg, let's give these researcher guys some research funds, 以防万一, never say never?(:)

10.

打战做生意要做的长久，料敌从宽是必须的 [ 山海马甲 ]

电脑病毒都有潜伏期，电脑木马、后门是否发作要看条件 [ 山海马甲 ]

电脑木马、后门要是没有激活条件，永远都不发作也是可能的。黑盒测试时无法涵盖这部分风险的，只有白盒测试才行。

基因后门同样也是这个道理，不是海南试种几代就可以的。 老兄是没接触什么安全工作吧？ 即时是中国人自己搞的，关键的事情也要有几个机构互相牵制，多个渠道和互相验证。这个道理古代帝王早就实施了。死生之道，自古以来都是慎之又慎的。

俺是搞软件的，计算机安全接触的也较多。 可以负责任的说，因为现有的基础硬件和软件目前还掌握在老美手里，我们在网上的每一个举动，老美只要想知道，就基本可以知道。 民用所有的加密、密码等措施，最多就是企业级、民用级的，对于美国国家层面的窃密，基本上就是裸奔的。 NSA如果预算足够，他很多时候比我们自己都知道自己。从基层的访问记录、发言记录、个人成长记录等，经过一定的统计分析和心理解析，就可以推断一个人的大致行为方式和思想去向。有时我们还没有开始判断，其就可以将我们对某事的判断预测的八九不离十了

obviously, a nation of 单色平面波 can hardly 料敌从宽;

obviously, 料敌从宽 with a 宽谱 mind has to be a national culture, 从小抓起, very likely a long "linear" process for a population or a nation, you have to kick the ass, bite your teeth, change mentality to get it started, then keep the momentum, can this ever happen to the Chinese society? if not, what is alternative? is there an alternative? simple logic stuff, but seems still very challenging to a country of 1.3B people of 5 k years of "风雨交加"?

11.

I do feel Chairman X is serouly lacking in his vision, aptitude, knowledge, ability & capacity in learning & understanding modern "war" of science, technology, including AI, IT, bio, life science, and of course, finance, capital market etc;

I also feel CAS (Chinese academy of science) have similar challenges they are facing

obviously, "mainland" china will be ok, regardless of whatever modern "war", in terms of keeping a reasonable economic growth rate (but TG does have to figure first how to unwind this Chinese red Ponzi game of world #1 size, then go from there), and as a result, X dynasty >10 years, but that is the so-called "ok", likely with not much variance on either direction.

10.

大国重器 (a very good series here in this forum, very informative), 识别区, etc stuff, of course, china should do and play those stuff, but as a country, or even a person in todays world, you are supposed to be able play multitasking game.

vs china's 单色平面波 mind as a nation:

重商主义, or 重器主义, but you have to do both, and you have to do at least three, with 重智主义 as well, play three or more balls in the air, or you are likely not a first class player;

and if you don't even have that concept or model in your mind, then regardless of your age, you will be always behind the curve, very likely

of course, if you don't care, nobody would care.