M. Fisz, Probability theory and Mathematical Statistics (Wiley, New York, 1963).  Home/Search   Document Not in Database   Summary   Related Articles   Check

....says that the e#cient estimate is one that has the least variance from the real parameter, and its variance is bounded below by the Cramer Rao bound. For example, the sample mean, w = i=1 w(i) is the e#cient estimate of the mean of a Gaussian distribution with a known standard deviation # [13]. For a vector version of the Cramer Rao bound, the reader is referred to [14, page 203 204] If we define w(i) u(i)u (i)x(i) v(n) in Eq. 2) can be viewed as the mean of samples w(i) That is exactly why our method is motivated by the statistical e#ciency in using averaging in Eq. 2) ....

M. Fisz, Probability theory and mathematical statistics, John Wiley & Sons, Inc., New York, third edition, 1963.

....at the kth time step. Then the position x i after i steps is given by: x i = x 0 v k 1T: 1) Since every v k follows the same but independent uniform distribution within the above velocity range, the distribution of x i can be approximated by a normal distribution (by central limit theorem [10]) the variance step of the motion added by one step is calculated as that of uniform distribution of width vmax 0 vmin , whichis(v max 0 v min ) 12. The probability density function p(x; i) of the obstacle being at x after moving for i steps is then given by p(x; i) p exp (x ....

M. Fisz. Probability Theory and Mathematical Statistics. Wiley, 1963.

....derivative does not exist everywhere. Since 1(8, does not depend on 0 and since f2 is bounded the rest of Condition 1, 60) is satisfied. We next verify Condition 2. The maximum likelihood estimates of a scalar valued parameter satisfy the Central Limit Theorem, provided Cramer s conditions, [4], on the differentiability of the likelihood function are satisfied. The proof extends to vector valued parameters provided the conditions hold componentwise. These condi tions require in the present case, first, that L, e) is three times differentiable in the interior of f2, which it is. ....

....to be verified. From (71) joc 3Z AYy 3dy e , e, and from (58) and (77) we get foe ) f(Y) 6be q 24ce 4y q e Sy2)dy (6bq 12c Hence the inequality (85) holds. Finally, the Fisher information matrix I(oz, e) is clearly bounded and positive definite, and by Cramer s conditions, [4], the Central Limit Theorem holds for the family p(ylx; To verify Condition 3 we get from (73) and (76) for Yt O xt 02L, yt O xt) while by (70) for Yt O xt e 1 02 t L, Yt O xt) n OOiO0j .oz(oz 1) Yt 2 , 0 Xt) XiX j O(ff( 2) 2O where the last inequality holds ....

Fisz, M. (1963), Probability Theory and Mathematical Statistics, John Wiley and Sons, Inc., New York, 677 pages

....does not exist everywhere. Since I( ffl) does not depend on and since Omega o is bounded the rest of Condition 1, 68) is satisfied. We next verify Condition 2. The maximum likelihood estimates of a scalar valued parameter satisfy the Central Limit Theorem, provided Cramer s conditions, [3], on the differentiability of the likelihood function are satisfied. The proof extends to vector valued parameters provided the conditions hold componentwise. These conditions require in the present case, first, that L ff;ffl (e) is three times differentiable in the interior of Omega Gamma which ....

Fisz, M. (1963), Probability Theory and Mathematical Statistics, John Wiley and Sons, Inc., New York, 677 pages

....l i = r i1 r i2 Delta Delta Delta r ik represent the load on server S i . The distribution of l i is approximately normal with mean k and variance koe 2 as N 1. Proof: The result follows immediately from the CentralLimit Theorem. See any standard book dealing with sampling theory, [10], for example. Hence, for sufficiently large packet train sizes, loads on all servers are normally distributed. We will now show that loads in a server cluster become balanced if the coefficient of variation (ratio of mean to standard deviation) of the load distribution on servers tends to zero. ....

....deviation, s= 0 as (oe= 0. Proof: Let l be the sample mean, and let y i = l i Gamma l. By the usual definition of sample variance, we have (m Gamma 1)s 2 oe 2 = i y 1 oe j 2 Delta Delta Delta i ym oe j 2 3 The sum on the right side follows the 2 distribution [10] with m Gamma 1 degrees of freedom 1 . Thus, we may write s 2 = oe 2 [ 2 m Gamma1 = m Gamma 1) Therefore, s= 2 = oe= 2 [ 2 m Gamma1 = m Gamma 1) Since the 2 distribution is bounded, oe= 2 [ 2 m Gamma1 = m Gamma 1) 0 as (oe= 0. Hence (s= 2 vanishes as ....

[Article contains additional citation context not shown here]

Marek Fisz. Probability Theory and Mathematical Statistics. John Wiley & sons, Inc., 1963.

....kth time step. Then the position x i after i steps is given by: x i = x 0 i X k=1 v k 1T: 1) Since every v k follows the same but independent uniform distribution within the above velocity range, the distribution of x i can be approximated by a normal distribution (by central limit theorem [7]) The variance oe 2 step of the movement added by one step is calculated as: oe 2 step = 1 vmax 0 vmin Z v max vmin (v 0 v) 2 dv = 1 12 (v max 0 vmin ) 2 ; 2) where v = v max v min ) 2 is the mean of the obstacle velocity. The probability density function p(x; i)of the ....

M. Fisz. Probability Theory and Mathematical Statistics. Wiley, 1963.

....drawn from some distribution. We consider two random distributions: ffl Uniform: The random variable has a uniform distribution, between [ Gammaff; ff] The mean of the random variable is 0. ffl Gaussian: The random variable has a normal distribution, with mean = 0 and standard deviation oe [Fis63] We fix the perturbation of an entity. Thus, it is not possible for snoopers to improve the estimates of the value of a field in a record by repeating queries [AW89] 2.1 Quantifying Privacy For quantifying privacy provided by a method, we use a measure based on how closely the original values ....

....Problem Given a cumulative distribution FY and the realizations of n iid random samples X 1 Y 1 ; X 2 Y 2 ; Xn Yn , estimate FX . Let the value of X i Y i be w i ( x i y i ) Note 3 that we do not have the individual values x i and y i , only their sum. We can use Bayes rule [Fis63] to estimate the posterior distribution function F 0 X1 (given that X 1 Y 1 = w 1 ) for X 1 , assuming we know the density functions fX and fY for X and Y respectively. F 0 X1 (a) j Z a Gamma1 f X1 (z j X 1 Y 1 = w 1 ) dz = Z a Gamma1 f X1 Y1 (w 1 j X 1 = z) f X1 (z) fX1 Y1 ....

Marek Fisz. Probability Theory and Mathematical Statistics. Wiley, 1963.

....13.9.15] r=n is the mode of the distribution. We can use the posterior distribution g, to find the probability that sample accuracy lies in any particular interval [a; b] P (p 2 [a; b] j p = r=n) 1 B(r 1; n Gamma r 1) Z b a p r (1 Gamma p) n Gammar dp (13) We have, for example [2] P (p 2 [0:9; 1]jp = 28=30) 1 B(29; 3) Z 1 0:9 p 28 (1 Gamma p) 2 dp = 0:611 (14) which gives a degree of belief of 0.611 that the rule s probability ( domain accuracy) is in the interval [0:9; 1] given that it has accuracy 28 30 and covers 30 examples in the training data. It is ....

Marek Fisz. Probability Theory and Mathematical Statistics. John Wiley, New York, third edition, 1963.

....of two well known sets of data, discrete and continuous, respectively. To calculate ae , we use some formulas put forward by Nelsen [14] for discrete data, and a numerical integration routine [16] to compute (4) for continuous data. 4.1 Discrete data M. Greenwood and G.U. Yule (see [5]) gave some examples of data following the negative binomial distribution (NB) rather than the Poisson distribution (PO) The data (DD) consists of the distribution of the number of accidents among 647 machine operators over several months . We consider five possible distributions for DD: fB; ....

M. Fisz. Probability theory and Mathematical Statistics. John Wiley & Sons, New York, 1963.

No context found.

M. Fisz, Probability theory and Mathematical Statistics (Wiley, New York, 1963).

No context found.

M. Fisz. Probability Theory and Mathematical Statistics. Wiley, 1963.

No context found.

M. Fisz (1961). Probability theory and mathematical statistics, New York.

No context found.

Marek Fisz. Probability Theory and Mathematical Statistics. John Wiley, New York, third edition, 1963.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC