r***y
发帖数: 25 | 1 :-) It's been THE debate ever since the simulation of neural network started
structure of the neural network in living creatures. Canonical AI people
said no, we just need to replicate the functions, which is faster but hard
to generalize. Neuroscience people said yes, we have to, otherwise we couldn
't replicate such complicated, decent, and flexible network, which makes the
question always stuck at the level of functioning units -- protein, synapse
, or single neuron? Any connection map seems ... 阅读全帖 |
|
r***y
发帖数: 25 | 2 :-) It's been THE debate ever since the simulation of neural network started
structure of the neural network in living creatures. Canonical AI people
said no, we just need to replicate the functions, which is faster but hard
to generalize. Neuroscience people said yes, we have to, otherwise we couldn
't replicate such complicated, decent, and flexible network, which makes the
question always stuck at the level of functioning units -- protein, synapse
, or single neuron? Any connection map seems ... 阅读全帖 |
|
f*******y
发帖数: 421 | 3 In parallel, we envision developing nanoscale neural probes that can locally
acquire, process, and store accumulated data. Networks of “intelligent”
nanosystems would be capable of providing specific responses to externally
applied signals, or to their own readings of brain activity.
这个nanoscale neural probes不是测电信号的吧？neuron network和intelligent的联
系有什么好文章？ |
|
f*******y
发帖数: 421 | 4 prefer background in mea, neural system modeling, neural circuit, etc.
please contact ylzhang at ntu.edu.sg for details. No pm. thanks. |
|
l*******s
发帖数: 1258 | 5 Neural network is so hard to interpret, and that's why some people don't
intend to use it in industry. When your client ask you to explain what
factors result in certain modeling conclusion, you can only tell them Neural
Network is not a regression... |
|
|
|
m***r
发帖数: 359 | 8 机器学习日报 2015-03-10
@好东西传送门 出品, 过刊见
http://ml.memect.com
订阅：给 [email protected]
/* */ 发封空信， 标题: 订阅机器学习日报
更好看的HTML版
http://ml.memect.com/archive/2015-03-10/short.html
1) 【如何评价季逸超、Peak Labs 和 Magi 搜索引擎？】 by @熊辰炎
关键词：知识工程, 季逸超, 知识库
我在 @知乎 回答了【如何评价季逸超、Peak Labs 和 Magi 搜索引擎？】：谢邀。 这
应该是在知乎最适合我的一道题目。趁着失眠来答一部分。 利益相关： 我是做知识图
谱和搜索的PhD，我之前一半的funding来自CMU机器学习系的never ending lang… [1]
[1] http://zhi.hu/gsnY
2) 【机器学习交互速查表】 by @爱可可-爱生活
关键词：资源, 课程, 速查卡
[速查] 机器学习交互速查表 [1] Scikit-Learn官网提供，在原有的Cheat Sheet基础
上加... 阅读全帖 |
|
m***r
发帖数: 359 | 9 机器学习日报 2015-03-14
@好东西传送门 出品, 过刊见
http://ml.memect.com
订阅：给 [email protected]
/* */ 发封空信， 标题: 订阅机器学习日报
更好看的HTML版
http://ml.memect.com/archive/2015-03-14/short.html
1) 【LINE模型大规模网络降维公布论文以及源代码】 by @chuckpku
关键词：应用, 社交网络
我们的论文“LINE: Large-scale Information Network Embedding”被WWW2015接收。
该论文提出的LINE模型能够在单机上把大规模网络降维到低维空间表示，稍后我们将公
布论文以及源代码 @张铭PKUCS
2) 【概率图模型基础总结】 by @数盟社区
关键词：算法
【概率图模型基础总结】在这一过程中，模型往往是必须的，什么样的模型才是相对正
确的？这又是我们需要解决的问题。这些问题出现在很多领域，包括模式识别、差错控
制编码等。概率图模型是解决这些问题的工具之一。 [1]
[1] http://dat... 阅读全帖 |
|
|
|
m***r
发帖数: 359 | 12 机器学习日报 2015-03-10
@好东西传送门 出品, 过刊见
http://ml.memect.com
订阅：给 [email protected]
/* */ 发封空信， 标题: 订阅机器学习日报
更好看的HTML版
http://ml.memect.com/archive/2015-03-10/short.html
1) 【如何评价季逸超、Peak Labs 和 Magi 搜索引擎？】 by @熊辰炎
关键词：知识工程, 季逸超, 知识库
我在 @知乎 回答了【如何评价季逸超、Peak Labs 和 Magi 搜索引擎？】：谢邀。 这
应该是在知乎最适合我的一道题目。趁着失眠来答一部分。 利益相关： 我是做知识图
谱和搜索的PhD，我之前一半的funding来自CMU机器学习系的never ending lang… [1]
[1] http://zhi.hu/gsnY
2) 【机器学习交互速查表】 by @爱可可-爱生活
关键词：资源, 课程, 速查卡
[速查] 机器学习交互速查表 [1] Scikit-Learn官网提供，在原有的Cheat Sheet基础
上加... 阅读全帖 |
|
m***r
发帖数: 359 | 13 机器学习日报 2015-03-14
@好东西传送门 出品, 过刊见
http://ml.memect.com
订阅：给 [email protected]
/* */ 发封空信， 标题: 订阅机器学习日报
更好看的HTML版
http://ml.memect.com/archive/2015-03-14/short.html
1) 【LINE模型大规模网络降维公布论文以及源代码】 by @chuckpku
关键词：应用, 社交网络
我们的论文“LINE: Large-scale Information Network Embedding”被WWW2015接收。
该论文提出的LINE模型能够在单机上把大规模网络降维到低维空间表示，稍后我们将公
布论文以及源代码 @张铭PKUCS
2) 【概率图模型基础总结】 by @数盟社区
关键词：算法
【概率图模型基础总结】在这一过程中，模型往往是必须的，什么样的模型才是相对正
确的？这又是我们需要解决的问题。这些问题出现在很多领域，包括模式识别、差错控
制编码等。概率图模型是解决这些问题的工具之一。 [1]
[1] http://dat... 阅读全帖 |
|
y*******n
发帖数: 148 | 14 【 以下文字转载自 CS 讨论区 】
发信人: yyqianyin (yyqianyin), 信区: CS
标 题: 审稿机会 Call for reviewers: soft computing 方向
发信站: BBS 未名空间站 (Thu Jun 11 01:14:51 2015, 美东)
2013 Impact Factor 1.304
大致方向：
soft computing technologies
evolutionary algorithms and genetic programming
neural science and neural net systems
fuzzy set theory and fuzzy systems
chaos theory and chaotic systems
有兴趣请发
* CV and/or homepage (including research interests and list of publications)
* Email
* Current affiliation
到 [email protected]... 阅读全帖 |
|
t******5
发帖数: 47 | 15 用recurrent neural network 和普通的neural network concatenate到一起就可以了
吧. |
|
t******5
发帖数: 47 | 16 用recurrent neural network 和普通的neural network concatenate到一起就可以了
吧. |
|
y******2
发帖数: 326 | 17 Linux是因为编程快，能用command line. 深层Neural network 几百万的parameter,
dot product用显卡快很多，因为内核多。游戏画多边形都是dot product。1070完全够
，有钱可以弄个1080ti。mnist 是最简单的，我平时都是＞50million x 500的矩阵，
1080得跑个两三天。neural network 其实是穷举，都是靠数据大，运算快，运算方法
并不先进。
4 |
|
w*****h
发帖数: 423 | 18 【 以下文字转载自 Programming 讨论区 】
发信人: waynezh (van), 信区: Programming
标 题: 有没有做sentiment analysis的，求思路
发信站: BBS 未名空间站 (Thu Oct 26 11:06:28 2017, 美东)
在公司做sentiment analysis，
现在的做法是这样
首先在大概几万句sentences中训练出word2vec model
然后将大概4000个labels的sentences转成vector, 80%作为training, 20%作为testing
， 用neural net作为classifier
类别三类，positive, neutral, negative
之前是公司的一个印度人在负责，达到63% accuracy
我接手之后调参 + 做一些ensemble提高到68% leader似乎不满足，觉得可以到70%甚
至80%
现在我想到思路有几个
1. 做semi-supervise, 没有label的数据有的是
2. unsupervise learning， 比如用autoe... 阅读全帖 |
|
s******a
发帖数: 38 | 19 Well, I agree with you that the implementation of
revolutionary interaction
method between human and computer might be extremely
difficult. But why do we
have to hold on to the idea of conventional computer? I
would envision some
kind of neural network with each person as a node, and also
lots of neural
computers as processing and storing units. A friend of mine
in biology is
studying the physical signs of monkey's brains' activity. He
made me believe
that the brain do produce different electro- |
|
h******w
发帖数: 332 | 20
Interesting idea! seems everybody is a fan of MATRIX! How
does security come in play with this neural network? The
current human society works as people expose the information
he wants to expose therefore realize the way of exchanging
whatever necessary. And the second issue is that with so
many electro-magnetic signals out there, how does the human
node filter/ignore garbage informations? The both issues may
be processed in the neural computer attached to the human
node. But the human node the |
|
c**i
发帖数: 6973 | 21 Michael Wines, A U.S.-China Odyssey: Building a Better Mouse Map. New York
Times, Jan 29, 2011.
http://www.nytimes.com/2011/01/29/world/asia
/29china.html?_r=1&scp=1&sq=fudan&st=cse
Note:
(a) Tian Xu, PhD, Department of Genetics, School of Medicine, Yale
Universoty, undated.
http://medicine.yale.edu
/genetics/people/tian_xu.profile
Quote:
"Professor of Genetics; Vice Chairman of Genetics; Investigator, Howard
Hughes Medical Institute; Director, Yale Center for Functional Genomics"
"Education[:]... 阅读全帖 |
|
|
|
|
m*********k
发帖数: 10521 | 25 =====================================================================
=========================Nature质疑叶诗文事件========================
=============================网友支招抗议============================
http://www.mitbbs.com/mitbbs_article_t.php?board=Biology&gid=31
发信人: zfwang (爱国老将), 信区: Biology
标 题: 对Nature 文章的反驳，作为千老，看不下去了。
发信站: BBS 未名空间站 (Fri Aug 3 10:24:23 2012, 美东)
原文是发在 Olympics 班 对给编辑写信的回复。但现在觉得这还是该生物班的责任。
作为一个科学家，一个中国人，有这么个争议性大的热门课题，放过可惜了。90% 的工
作量是收集数据，但人多力量大，如果后来的收集数据多了，可以裁开发，力度更大。
－－－－－... 阅读全帖 |
|
d********r
发帖数: 303 | 26 I am not proposing independence now.
But we should be prepared and if something happens, say CCP collapsed and
China is in turmoil, we should save our hometown by declaring Wu-Yue
independence.
Since now it is not advantagous for Wu-Yue independence, I obviously do not
support other non Wu-Yue provinces independence. I am neural to Tai Wai
independence. But I am extremely against Xing Jiang independence, because
Wu-yue area need Xing Jiang's gas (actually central asia's gas, but Xing Jiang
is th |
|
T*******x
发帖数: 8565 | 27 【 以下文字转载自 Programming 讨论区 】
发信人: wdong (万事休), 信区: Programming
标 题: 我来讨论下意识的问题吧
发信站: BBS 未名空间站 (Mon May 29 17:11:29 2017, 美东)
坛子里很多人喜欢提意识，甚至意识的量子机制，
其实本质上来说是用一些说不清的东西蹚浑水。
我觉得很多人其实没有认真思考过这些问题。
这些东西我思考过很多，而且还在不断思考，
下面说说我的看法。
佛教里有八识之说。其中前五识眼耳鼻舌身，其实
就是目前普通计算机能处理的视觉，听觉等输入。
第六识意识，就是神经网络的自反馈。从实现上来说，
就是把部分神经元的状态作为神经网络的输入，也就是
recurrent neural network。
佛教的第七识末那识，也就是“我执”的起源，
我觉得还可以再分解。其中的“执”，其实就是
AI的优化目标。对于人类而言，这个目标是写死在
基因里的。对于AI而言，则是程序员指定的一个目标。
以上这些，目前AI理论中都已有对应部分。
至于“我执”中的“我”，这个才是坛子里的诸位
在谈“意识”时真正所指的东西。如果... 阅读全帖 |
|
c******3
发帖数: 194 | 28 【 以下文字转载自 Biology 讨论区 】
发信人: FatSheep (Serta is my cousin), 信区: Biology
标 题: 美国变态PI列表 转载
发信站: BBS 未名空间站 (Sun Apr 18 19:53:07 2010, 美东)
http://bbs.bio668.com/read.php?tid=35969
美国变态PI
David X. Liu Department of Neural and Behavioral Science Penn State Univ. 2
postdoc were fired after working for Dr. Liu only for 2 months because they
did not obtain the expected experimental results.
Roger (Yuguang) Shi Department of Cellular & Molecular Physiology Penn State
Univ.
was in Department of Biochemis |
|
l**i
发帖数: 8144 | 29 【 以下文字转载自 Biology 讨论区 】
发信人: jurassicpark (猪拉稀·怕渴), 信区: Biology
标 题: 整理了“美国变态PI列表”，欢迎转载到国内各高校bbs
发信站: BBS 未名空间站 (Wed Apr 21 14:28:38 2010, 美东)
发信人: FatSheep (Serta is my cousin), 信区: Biology
标 题: 美国变态PI列表 转载
发信站: BBS 未名空间站 (Sun Apr 18 19:53:07 2010, 美东)
http://bbs.bio668.com/read.php?tid=35969
美国变态PI
David X. Liu Department of Neural and Behavioral Science Penn State Univ. 2
postdoc were fired after working for Dr. Liu only for 2 months because they
did not obtain the expected experimental re |
|
s********n
发帖数: 26222 | 30 This is such a a stupid argument. Let's say it might be true that South
Korea's per capita GDP is $17,708. However, this GDP is produced by the
South Korean people. This number has nothing to do with Americans and US
government. How could you attribute this GDP number to the Korean war? This
is the most absurd logic I have ever heard of. You brain might be dying and
you should think of neural stem cell therapy seriously.
GDP
calculated. |
|
s********n
发帖数: 26222 | 31 你们自己看一下。 这个网站貌似是香港人或者台湾人搞的网站。
http://multipletext.com/2010/9_deputy_sheriff.htm
The following is the main point of an article penned by the chief Beijing
correspondent for Morning News (新闻晨报驻京首席记者):
There was a time when I began to wonder from where Fang got money to support
himself and his family. Later Professor Xiao Chuanguo (肖传国教授) launched
defamation case against Fang in a US court and asked his US lawyer to
investigate Fang's website that is largely used to publically smear anyone
and anyt... 阅读全帖 |
|
p*****c
发帖数: 20445 | 32 Science评价还是不错的，专门写了News：
Science 5 November 2010:
Vol. 330. no. 6005, p. 747
DOI: 10.1126/science.330.6005.747
News of the Week
Neuroscience:
China's Brain Mappers Zoom In on Neural Connections
Kristen Minogue
Anyone in business, and even just fans of The Apprentice, knows that
connections matter. Neuroscientists understand that, too, which is why a
research team in China has gone to great lengths to create the most detailed
three-dimensional map yet of all the connections between the neurons in... 阅读全帖 |
|
e*****s
发帖数: 7359 | 33 I think if you have some knowledge of biology, evolution, and neuroscience,
you would probably agree with him.
the difference between human and other primate mammals such as apes is not
that big in terms of neural hardware. it's very likely largely due to the
difference in computing capacity. Besides, intelligence does not need the
prerequisite of reproduction. |
|
s**********l
发帖数: 8966 | 34 呵呵，我也做过EEG，fMRI，还专门做neural activity的feature extraction，啥
fancy方法，啥L1
Optimization啊，Variational Bayesian分层extraction之类都实验，斯坦福一帮做的
最好的人也只能
根据preset的model detect喝水动手这类的活动，打个游戏之类稍微难点的脑电波信号
就完全找不到
啥correlation了。上面那位的蛋到底是从何扯起啊？
先问个简单的问题吧，他要是懂得话，现在EEG最大分辨率多少啊，一般业界分析EEG
signal用啥软件
啊？ |
|
x****u
发帖数: 12955 | 35
用neural network 的 AI， 可以通过大量模拟来得到一个类似的概念。 |
|
m*********r
发帖数: 1797 | 36 neural network?
关键是如何选取数据和多少层的神经元，
训练这么牛鼻的机器人，
不是这么简单的。 |
|
|
|
|
s*******u
发帖数: 9508 | 40 下面是所有editor的信息，但是没有email地址。 应该能查出来。
大家给这些editor发信吧，comments的作用不是很大。
ABOUT THE EDITORS
Editor-in-Chief
Philip Campbell
London
Education: BSc, aeronautical engineering, University of Bristol; MSc,
astrophysics, Queen Mary and Westfield College, University of London; PhD
and postdoctoral fellowship, upper atmospheric physics, University of
Leicester. Areas of responsibility include: Co-Editor of Editorials,
editorial content and management of Nature, long-term quality of all Nature
Publications.
Ex... 阅读全帖 |
|
l**1
发帖数: 1875 | 41 是教授
宋森 博士、研究员、博士生导师
1996年本科毕业于密西西比大学，2002年获该布兰德斯大学神经学博士学位。2001-
2002年在GPC生物公司任计算生物学家。2002-2004年在冷泉港实验室任博士后。2004-
2010年在MIT继续博士后研究。2010年聘为清华大学医学院生物医学工程系研究员，目
前是Physical Review Letters, Neural Computation, Frontiers in Neurosciences,
PLOS Computational Biology等杂志审稿人。
联系地址：清华大学医学科学楼 生物医学工程系 100084
联系电话：+86-10-62773357
传真：+86-10--62773380
电子邮件： s******[email protected]
Sen Song Ph.D.
Professor
Department of Biomedical Engineering |
|
P*****t
发帖数: 4978 | 42 Inverse reinforcement leaning, not neural network. |
|
x****u
发帖数: 12955 | 43
It's probably just how each person's brain process neural signals. In
essence, it's a software function, not the hardware. Unlike a computer, you
can't step through the code to see how the signal is processed. |
|
t*****9
发帖数: 10416 | 44 都去搞 Neural Networks了，憋屈呀 ～～ |
|
f******i
发帖数: 4647 | 45 搞machine learning的人都知道，unbiase的模型虽然好，但是往往很难得到。
这个时候掺点假，容忍一点错误，结果还凑合，大家也能接受。
在公司里面混也是这样，索男们要多学学machine learning的现代算法，获得一些人生
启迪。
machine learning里面另外一个重要原则，是在exact solution很难得到的情况下，
approximation也能管用。所以做人不要那么认真。阿三就是精通了machine learning
在职场上的运用。
另外，machine learning里面的比如neural network这样的算法，很难解释，但是他妈
的就是管用。索男们应该时时提醒自己，很多时候，管用就行，别想那么多。 |
|
d*******3
发帖数: 8598 | 46 你这个土人，80年代就有machine learning，neural network了
这玩意太难，以至于不能成为工程产品
你要能搞出来AI，你就是可以和牛顿高斯平级的人物 |
|
d*******3
发帖数: 8598 | 47 particular brain function.
The structure complexity of the brain
imposes a formidable task for deciphering
the neuronal basis of brain functions
at themicroscopic level.Thehuman brain
consists of 100 billion neurons of hundreds
of different types, with 1015 specific
connections among them that form
neural networks underlying diverse brain
functions.The best current technology allows
simultaneous monitoring of activities
from about a 1000 neurons. Scaling
up the number of neurons by a few orders
... 阅读全帖 |
|
g****5
发帖数: 1639 | 48 没有必要望文生义地认为现阶段的人工智能就真的是想模拟高级生物比如人的智能。就
当成是“夫妻肺片”这样的概念理解就行了。人工智能只是个学术界忽悠出来用以骗经
费的概念而已。顶多相信后面俩字“智能”就行了，跟“人工”实在没关系。其实还是
machine learning这个词比较贴切，就是让silicon machine具有一定学习的能力，做
自动的决策。
另外搞生物的也有误解，认为人工智能是去试图模拟人脑的智能。从现阶段人脑的工作
模式从生物上都没搞清楚来攻击，这也是没必要。还是那句话，跟人工无关，只是利用
现有的计算条件，做一些简单的学习模型和预测。你们去搜一搜，AI的一个经典定义是
这样的：The use of computers to solve problems that previously could only be
solved by applying human intelligence. 按照这个定义，连qq的棋牌室都算，因为
先前只有人才会下棋。另外一个定义则比较接近现在的机器学习：The use of a
specific set of programming te... 阅读全帖 |
|
b*****n
发帖数: 240 | 49 ...discovering two types of neurons has nothing to do with neural network...
that's just another seemingly related thing that people like to pick up with
... |
|
b********n
发帖数: 38600 | 50 a neural network that approximates a global brain—is fusing with the human
mind in a phenomenon that is called “singularity,” and they’ve hired
transhumanist scientist Ray Kurzweil to do just that. Google will know the
answer to your question before you have asked it, Kurzweil said. “It will
have read every email you will ever have written, every document, every idle
thought you’ve ever tapped into a search-engine box. It will know you
better than your intimate partner does. Better, perhaps, tha... 阅读全帖 |
|
