t******g
发帖数: 10390 | 1 100个动物冷知识
1、“hoka-hoka”是倭黑猩猩（Pan paniscus）百合时特有的行为，两只倭黑猩猩女生
紧紧贴在一起，然后用力摩擦敏感部位。
2、花喜鹊是唯一能从镜子里认出自己的非哺乳动物。
3、树袋熊一天要睡18个小时。
4、达尔文吃过美洲狮的肉，他说味道跟小牛肉一样。
5、气步甲（Brachinus favicollis）喷出的毒液温度可达100℃。
6、蓝鲸的喉咙可以吞下一只沙滩排球。
7、雄蜘蛛没有JJ，它把精液排在一张小蛛网上，用嘴边的触角像灌钢笔水一样吸进去
，给雌蜘蛛注射。
8、有一种寄生蝇（Ormia ochracea）的耳朵长在胸前，它对蟋蟀求偶的声音特别敏感
，因为它的卵产在蟋蟀体内。
9、涡虫的脑袋和尾巴砍掉都可以再生，而且再生脑袋的速度比再生尾巴快。
10、大王乌贼没有攻击人的可靠记录。
11、母非洲水雉（Actophilornis africana）有后宫，一般由四只公水雉组成，她保卫
领地，而他们负责孵蛋。
12、花斑鹟（Ficedula hypoleuca）是一夫一妻的鸟类，但有39%的雄花斑鹟在老婆之
外有外遇。
13、母猪前面的奶头比后... 阅读全帖 |
|
s********n
发帖数: 4346 | 2 What proposed here is a Bayesian model of predicting whether a particular
person is P, H or T.
Let Type denote a discreet variable of predictions P, H or T.
Let s denote an observed variable of appearance or attribute.
P(Type) is the probability of T, P, H based on common sense and statistics
without knowing anything about the particular person.
P(s,Type) is the probability of seeing characteristics s (long hair, shorts
, red Jeep, worrying about marrying a man, etc) given the Type.
Bayesian i... 阅读全帖 |
|
k****u
发帖数: 1686 | 3 Un mentore comunista benedetto Noè e di suo figlio, disse loro: "Siete
chiamati a sopportare e posteriore, un grande Bianman della terra.
2 Se le bestie della terra e l'aria sono tutti i volatili devono avere paura
, paura di te, anche sul terreno di tutti gli insetti, e tutti al mare
Pesci sono consegnati a mano.
3 Se gli animali viventi possono essere utilizzati per il vostro cibo. Tutto
questo io ti do, la stessa verdura.
4, tranne la carne con il sangue, che è la sua vita, non si può mangiar |
|
k****u
发帖数: 1686 | 4 Un mentore comunista benedetto Noè e di suo figlio, disse loro: "Siete
chiamati a sopportare e posteriore, un grande Bianman terra.
2 Se le bestie della terra e l'aria sono tutti i volatili devono avere paura
, paura di te, anche sul terreno di tutti gli insetti, e tutti al mare
Pesci sono consegnati a mano.
3 Se gli animali viventi possono essere utilizzati per il vostro cibo. Tutto
questo io ti do, la stessa verdura.
4, tranne la carne con il sangue, che è la sua vita, non si può mangiare.
5 f |
|
c****t
发帖数: 19049 | 5 THE GHASTLY ORDEAL OF TIMOTHY CAVENDISH
One bright dusk, four, five, no, my God, six summers ago, I strolled along a
Greenwich
avenue of mature chestnuts and mock oranges in a state of grace. Those
Regency residences
number among London’s costliest properties, but should you ever inherit one
, dear Reader, sell
it, don’t live in it. Houses like these secrete some dark sorcery that
transforms their owners into
fruitcakes. One such victim, an ex-chief of Rhodesian police, had, on the
evening in qu... 阅读全帖 |
|
H*U
发帖数: 95 | 6 俺来试试俺对什么是科学的初步理解。主要参考是俺前面贴的几个链接。欢迎拍砖！：
-）大家一起讨论。
科学这个词的意思是知识/认知。是人类通过人类掌握的手段（科学方法）对人类能够
感知的周围环境（人类生活的四维空间世界）的， 能够被人类感知，证实的人为假设
。 科学以及知识/认知是一个范畴， 也是一个动态的概念。
Aristotle 在《Posterior Analytics》中对知识/认知给出的理解是：We suppose
ourselves to possess unqualified scientific knowledge of a thing, as opposed
to knowing it in the accidental way in which the sophist knows, when we
think that we know the cause on which the fact depends, as the cause of that
fact and of no other, and, further, that the fact could not b... 阅读全帖 |
|
l**********t
发帖数: 5754 | 7
---后验的上帝? can you clarify since I'm not sure about this term. do you
mean posterior, or shall it be ex ante predictability?
子在下落过程中和在地上的随机碰撞
-- so you agree that something may seem random to you and me but can be
totally deterministic from God's perspective?
it is not his definition, it is THE definition from the Bible --预知未来的
全
能个体, because "time" doesn't mean anything to God because he exists
beyond
the constraints of space & time.
Rev 1:8 I am Alpha and Omega, the beginning and the ending... 阅读全帖 |
|
E*******1
发帖数: 3464 | 8 问题是，
1.量子力学本身是概率统计的物理推论，根基在于缺乏对一些信息的了解，所谓的
prior robust和posterior robust模型。由于测量只能根据已有的信息构建全部信息，
必然引入部分未知参量，比如量子的波性质就是这样而来。担不是说这个是它的本质属
性，而只是测量的信息不完全的属性而已。简单的例子，仍骰子，是一个概率问题吧，
但是源于什么呢？源于我们对于每一次扔的初始条件等缺乏完全的掌握，如果我们知道
对色子的初始速度和力的准确把握，这不过是一个经典的刚体转动问题，完全不存在什
么50%正反概率问题，而可以准确解出其值。
2.量子力学强调测量，是对某一时间，需要考量其对本征态的分布而言，但前提是肯定
了本征态和本征值的概念，且本征值是时间的连续函数。如果力学量和哈密顿对易，即
量子力学的守恒量，其测量值几率是不含时间的定态分布。比如，电子，不过是处在动
量本征态，这个本身是和测量与否无关的。
总之，你如果拿量子力学说事，那首先得承认本征态的概念，而本征态并不是和测量同
时存在的。如果说只有测量时候才有态这个概念，那你还是彻底否认了物理的基本原理 |
|
d**i
发帖数: 9682 | 9 wtf is that... no need to show us your preference of anterior positions over
posterior ones. |
|
f******k
发帖数: 297 | 10 what does kalman filter have anything to do with bayesian framework? it
neither has any prior settings nor calculate any posteriors... |
|
o****e
发帖数: 92 | 11 prior is supplied by your model. posterior is after training. |
|
s******e
发帖数: 285 | 12 看Carl的这本书，讲的很仔细。
http://www.gaussianprocess.org/gpml/
简单的来说就是个stochastic process, assuming a
GP prior.
一般的算法是假设input x是normal distribution,
GP是假设function space f(x)是normal distribution.
对于regression来说，因为y的值是连续的，所以
likelihood function p(y|f(x))也可以表示为Gaussian
的形式，这样一来posterior p(y|x) = \int p(y|f(x)) p(f|x)
就也是guassian的了，due to conjugation.
Classification的情况复杂一些，需要一个 multi-logic
function来转换p(y|f(x)).
Christ Bishop的书也写得很浅显易懂。 |
|
D*******e
发帖数: 258 | 13 如果确定了model是Gaussian，mu和sigma2就是平均值和方差
没记错的话ML的结果有偏，是 sum((x-x_mean)^2)/N，不是除以N-1，好像要用bayes的
posterior修正，等于加一个参数方程复杂度的penalty，得到的结果就是sigma2等于方
差了。
查查statistical learning之类的吧
如果你要加weight，再让weight adaptive，那相当于adaptive base function，查查
bishop那些人的书看看吧。 |
|
K****n
发帖数: 5970 | 14 和一个学math的人讨论着讨论着有点儿糊涂
条件概率 P(A|B)，如果写出这个符号，是否说明B一定是个常数，不能是个变量。。。
我怎么觉得不用
是常数
那 P(A}B)P(B)=P(B|A)P(A)这个公式怎么说得通，岂不是A和B都是常数？
要是做Bayesian regression, w 是要估算的参数，D是收集到的数据，想maximize
posterior
P(w|D), 那写要maximize P(D|w)p(w)有问题么 |
|
m**s
发帖数: 346 | 15 【 以下文字转载自 Statistics 讨论区 】
发信人: meds (忘记吃药了！), 信区: Statistics
标 题: 付费找tutor
发信站: BBS 未名空间站 (Sun May 29 21:52:33 2011, 美东)
修一门bayesian methods in computer science。
主要是编程实现一些统计上面的算法。
开始还好，最近的一个lab，让我很头大。也许给我很多时间，我能想出来，但是会太
迟了。明天就要交了，虽然是memorial day。
不是想找人帮忙做，只需要懂的人看看，点拨一下。为了表示对您的时间的尊重，我愿
意$10/hr 补偿您的时间。钱肯定不多，您如果嫌不够，可以再商量。
下面是这个lab的大概意思：
implement MCMC using Metropolis (or Metropolis-Hastings).lab给了4个bayesian
networks，要用我们写的程序针对某个query，做sampling，然后plot mixing, prior,
and posterior. 其实我自己感觉这个lab有些地... 阅读全帖 |
|
W***o
发帖数: 6519 | 16 不是很明白
现在的Chrome默认设置就是显示你最近关闭的tab的 snapshot，你意思是说要放大看那
个snapshot? 这有啥看头儿，不如访问一下阅读了
还有，那个不应该叫preview吧，因为是以前的，应该叫 posterior view 或者
snapshot? |
|
a*****t
发帖数: 81 | 17
greece 看来你真的对stroke很感兴趣， 原来上neuroscience时候老师讲了一些
关于stroke(中风)的病例分析， 但对脑部血管的分布基本上都快忘记光了， 不过
你去图书馆找本Human Atalas的书都可以翻到的。 我所记得的一些是， stroke
泛指由于脑血管疾病造成的脑部功能障碍， 大致有两种， 脑部血管破裂， 受损害
区域很大， 临床症状也很复杂; 另外一种是局部血管阻塞， 受损害区域较小，
具体表现为某种或数种感觉或运动功能丧失。 象你说得病例，如果 只是 occipital(枕)
lobe血管阻塞， 由于occipital lobe 又被称作 visual cortex， 可能的症状就是视觉
功能受损， 供应occipital lobe的血管是 posterior cerebral artery, 同时此血管还
供应一部分ventral cerebrum surface area的血液， 任何有关temporal lobe的症状
也 都有可能发生， 象听觉，记忆之类. 我觉得如果不存在cervical desease的 |
|
f***4
发帖数: 886 | 18 谢谢回复
那活体上直接做的（就是说不是细胞水平的），意义大吗。
看这片文献http://www.sciencemag.org/cgi/data/309/5743/2033/DC1/1
Electrophysiology. Tc1 and wild-type littermate mice, aged 4-8 months, were
anaesthetized with an intraperitoneal injection of urethane (1.8 g/kg). LTP
was investigated in the dentate gyrus using a concentric bipolar stimulating
electrode positioned in the perforant path (AP-level with Lamda, ML-3 mm
from midline, DV-1.5 mm) and a glass micropipette recording electrode
positioned in the hilus of the ipsil... 阅读全帖 |
|
w********y
发帖数: 288 | 19 意义大不大跟具体研究的问题有关吧。活体实验一般能把生理机制跟行为功能直接联系
起来，笼统的讲比脑片意义大。反之，如果不直接研究生理功能，就简单的做一些电生
理数据测量，即便实验难度大，意义也有限。
were
LTP
stimulating
posterior
by |
|
l*******8
发帖数: 1745 | 20 heart failure: 左心衰还是右心衰。diastolic还是systolic? chamber有没有增大？
左心衰除了left anterior wall, left posterior wall会有改变外，septum有时也会
有改变吧？
如果slide上能反映出这些会不会更好一些？有相关文献可以参考吗？ |
|
d*****u
发帖数: 17243 | 21 数据量大，按照law of large numbers，样本均值会逼近随机变量的期望
按照central limit theorem，样本均值会逼近成正态分布
但有些统计问题并不是研究期望的，也不关心正态分布
比如研究某种刺激对人的行为的影响
一次刺激是一种影响，两次刺激又是另一种影响
统计学（特别是bayes学派）也研究posterior distribution的这些变化 |
|
s******s
发帖数: 13035 | 22 我就不说sampling那些好玩的东西了。就算bayesian第一门课就要学的
posterior = prior x likelihood， 这玩意儿念过prob101就会推导的
东西，完全是博大精深啊：小到对parameter的理解，大到科学方法论和
对世界的理解。又比如作为理科生怎么理解racial profile这类问题。
贝叶斯强就强在这个基石理论连高中生都能推导的出。 |
|
s******s
发帖数: 13035 | 23 我就不说sampling那些好玩的东西了。就算bayesian第一门课就要学的
posterior = prior x likelihood， 这玩意儿念过prob101就会推导的
东西，完全是博大精深啊：小到对parameter的理解，大到科学方法论和
对世界的理解。又比如作为理科生怎么理解racial profile这类问题。
贝叶斯强就强在这个基石理论连高中生都能推导的出。 |
|
|
l**********1
发帖数: 5204 | 25 生物统计的入门教材之一 哈 of course english version:
http://www.roma1.infn.it/~dagos/rpp/
Introduction
Uncertainty and probability
Rules of probability
Probability of simple propositions
Probability of complete classes
Probability rules for uncertain variables
Bayesian inference for simple problems
Background information
Bayes' theorem
Inference for simple hypotheses
Inferring numerical values of physics quantities -- General ideas and basic
examples
Bayesian inference on uncertain variables and posterior ... 阅读全帖 |
|
|
N******n
发帖数: 3003 | 27 the usual Bayesian approach obtains the posterior probability through MCMC
sampling and takes longer time. ML gets its parameter estimation through
numerical optimization and therefore, faster than BI.
bootstraps |
|
r******0
发帖数: 357 | 28 BMC终于在最近做了一件值得尊敬的事情。我敢说下面的狗屎99%都是所谓的个性化生物
“信息”“润色”的畸形怪胎。一般以所谓的Meta分析或者随便下载一套狗屎基因芯片
数据，然后做一些似是而非的烂分析，发一堆垃圾，一篇文章卖给忙于召妓的医生5万
到10万。
1.
Retraction Note Open Access
Retraction Note: Impact of guidewire selection and operator expertise on
radiation exposure in transradial angiography
Jianmin Yang, Ningfu Wang, Xiaoshan Tong, Xianhua Ye, Liang Zhou, Guoxin
Tong, Yun Shen, Shuzheng Lv
Journal of Cardiothoracic Surgery 2015, 10:42 (27 March 2015)
Full text | PDF
2.
Retraction Note Open Access
Retracti... 阅读全帖 |
|
v********a
发帖数: 646 | 29
哪篇文章啊，谢谢
Rinn JL, M. Kertesz, J. Wang, S. Brunghman, H. Goodnough, B. Wong, M. Cleary
, J.A. Helms, E. Segal, H.Y. Chang. (2007) Functional Demarcation of Active
and Silent Chromatin Domains in Human HOX Loci by Non-Coding RNAs. Cell. Jun
29;129(7):1311-23.
F. Lan, P.E. Bayliss, J.L. Rinn, J.R. Whetstine, J.K. Wang, S. Chen, S.
Iwase, T.M. Roberts, H.Y. Chang, and Y. Shi (2007). A histone H3 lysine 27
demethylase regulates animal posterior development. Nature Oct 11;449(7163):
689-94 (online 9... 阅读全帖 |
|
d********m
发帖数: 3662 | 30 thanks.
不知道你说指的说反了是什么。定义了loss function，然后能做的无非就是1，
optimize，比如regression smoother，2，posterior expectation。model的好坏不就
都反应在这里面了？统计基本上所有的问题都是optimization，要么数值，要么EM，要
么MCMC。
另外，能具体展开说一下deep learning的核心code大概是指什么？我现在在做一个
gaussian mixture model的parameter estimation，写一个gibbs sampler核心都不止
十几行。不过你对coding那段的确对我启发很大，我coding的确是
不行，可能对你所说的理解不到位。 |
|
d********m
发帖数: 3662 | 31 sorry I am sitting with the lab desktop so I can't type Chinese.
but again, in the simplest case, the general theory for nonparametric
smoother is
all about prediction. summarization of predictive posterior distribution
with, for example, monte carlo is again all about prediction. I fail to see
how the line you drew between statistics and ML is relevant. |
|
J**Y
发帖数: 34 | 32 In Bayesian econometrics, we need estimate the moments of functions of
interests, for example, the expecttation of f(a): E[f(a)|D], where
a is a random number and D is observed data.
E[f(a)|D]=Integrate[(f(a)*p(a|d)),{a}]. Where p(a|D) is posterior
density of a. Generally, we can not find
the closed form of the integration, so some numeriacl methods is neccessary.
The first method is Laplace's method. This method changes the problem
to a numerical derivative, which is much easier. However, this |
|
t******s
发帖数: 39 | 33 To idioteque:
握手。同是宏观经济学生。
用Bayesian方法估计新凯恩斯主义DSGE模型时，对于prior distribution的设定和weak
identiciation的问题的确有争论，但这不能否定Bayesian估计方法，它在理论上是站
得住脚的。在我们所关心的dimension上，prior standard error自然设定大些。如果
tight prior对于posterior估计结果和marginal data density影响很大，那这样得到
的结果自然不robust,负责任的研究不会这样做，但tight priors在有些时候是可以接
受的。0.0001的prior standard error不常见。
“我的观点是，宏观经济学研究者（包括我自己）所了解和掌握的宏观经济规律不多，
现在的方法存在很多问题，还是承认我们的无知更好。” |
|
X*********e
发帖数: 253 | 34 no, its depend on what is your posterior distribution. |
|
e****c
发帖数: 183 | 35 Looks like there are two different scenarios. One is that you have the same
information before and after you bought the car, or you have some new
information after the purchase, but the new information either corroborates
with your purchase decision, or is not detrimental enough as to make you
think you had made a wrong purchase. I wouldn't call this as posterior
justification, since you still honestly think you've made a right decision.
The other situation is that you've got some new informatio |
|
h**********a
发帖数: 40 | 36 No. 1: First and Second Law of Macroeconomics:
1) Linearity of Logarithm: log (A+B) = log (A) + log (B)
2) Jensen's Equaliy: E[f(X)] = f(E[X])
No. 2: Why does a lady refuse to marry a Bayesian statistician?
Because he only cares about her *posterior* distribution. |
|
I***e
发帖数: 1136 | 37 你的推理并不正确.
X is a R.V., X+1 is also an R.V. Conditional on you observed 3^N dollars you d
o not
know for sure what X is right? And there is a perfect probability distribution
.
To further illustrate, let's say you do the following many many times:
1. Pick a X
2. 2 envelopes one with 3^X, one with 3^(X+1)
3. Pick one at random, give it to you and you open it.
Now let's say that you have observed $27 in it: there must be a posterior dist
ribution
of X right? I am telling you that P(X=3|observed 27) |
|
G*****7
发帖数: 1759 | 38 naive bayes的posterior无穷大？
描述一下你运算贝叶斯模型的过程是指什么 |
|
B****n
发帖数: 11290 | 39 一個簡單的方法就是先主觀的定它的難度(依據需要思考的步數 分支 有沒有需要"妙手
"等)
然後呢把它放線上由使用者來玩 按照答對的程度 來adjust
這就屬於Bayesian model 先有一個prior distribution(主觀難度) 再由使用者答題狀
況來update你的distribution 得到所謂的posterior distribution 很多線上排序的東
西就是用Bayesian model來弄得 |
|
l*********e
发帖数: 158 | 40 I guess; Sitting - Deep Venous Thrombosis in the lower limb -Patent foramen
ovale- stroke in deep penetrating arteries that arises directly from the
constituents of the Circle of Willis, cerebellar arteries, and basilar
artery that irrigate posterior limb of the internal capsule-descending
corticospinal tracts- hemiplegia? How can the bilateral lesions happen with
one side of symptoms? |
|
a****9
发帖数: 773 | 41 A 5-month-old boy is brought to the physician because of a 24-hour history
of fever, cough, noisy breathing, and difficulty feeding. His symptoms began
3 days ago with nasal discharge, mild cough, and chest congestion. He
appears somewhat irritable and is crying. His temperature is 38.5 C (101.3 F
), pulse is 108/min, and respirations are 32/min and shallow with a
prolonged expiratory phase. On examination, the throat appears normal. A few
small anterior and posterior cervical nodes are palpable |
|
a****9
发帖数: 773 | 42 14. A 27-year-old primigravid woman at 12 weeks' gestation comes to the
emergency department 2 hours after the sudden onset of bright red vaginal
bleeding. She has not had abdominal cramping. Pelvic examination shows a
small amount of brownish blood in the posterior fornix of the vagina. The
cervix is closed. The uterus is palpable 3 cm above the pelvic brim. Fetal
heart tones are easily audible at 167/min by Doppler. Which of the following
is the most likely diagnosis?
A) Abruptio placentae
B) |
|
a****9
发帖数: 773 | 43 A 5-month-old boy is brought to the physician because of a 24-hour history
of fever, cough, noisy breathing, and difficulty feeding. His symptoms began
3 days ago with nasal discharge, mild cough, and chest congestion. He
appears somewhat irritable and is crying. His temperature is 38.5 C (101.3 F
), pulse is 108/min, and respirations are 32/min and shallow with a
prolonged expiratory phase. On examination, the throat appears normal. A few
small anterior and posterior cervical nodes are palpable |
|
a****9
发帖数: 773 | 44 A few small anterior and posterior cervical nodes are palpable.
I thought this is a hint for adenovirus ? Correct me if i am wrong.Thanks! |
|
l**********e
发帖数: 382 | 45 拍X-ray了吗？anterior and posterior drawer test 是用于协助诊断ACL, PCL损伤. |
|
