In a previous work (S. Fiori, 2006), we proposed a random number generator based on a tunable non-linear neural system, whose learning rule is designed on the basis of a cardinal equation from statistics and whose implementation is based on look-up tables (LUTs). The aim of the present manuscript is to improve the above-mentioned random number generation method by changing the learning principle, while retaining the efficient LUT-based implementation. The new method proposed here proves easier to implement and relaxes some previous limitations.
1. Introduction
Random numbers are currently used for a variety of purposes such as:
cryptographic keys generation, games, some classes of scientific experiments
as well as “Monte Carlo” methods in physics and computer science
[1–6].
Standard programming environments are endowed with basic pseudorandom signal
generators such as the uniform and the Gaussian ones, while usually the needed
distributions are more involved than uniform/Gaussian. A simple example of
application is password generation: a random password generator is a software
that inputs from a random or pseudorandom number generator and automatically
generates a password. An example of application where involved probability
distributions are needed is in independent component analysis (ICA, [7]) testing: as
the behavior of an ICA algorithm might depend on the statistical distribution of
the sources, ICA-algorithm testing tools might require random sequences generators
capable of producing random numbers distributed according to involved probability
laws.
The principal methods known in the literature to
obtain a batch of samples endowed with an arbitrary distribution from a samples
batch having a simple distribution are the “transformation method” and the
“rejection method” [8]. In the
present paper, we focus on the transformation
method, which may be well implemented through a tunable neural system, because
the availability of a random number source and of a tunable nonlinear system,
along with a proper learning procedure, allows obtaining a wide class of
pseudorandom signal generators.
A well-known effect of nonlinear neural systems is to
warp the statistical distribution of its input. In particular, we assume that
the system under consideration has a nonlinear adaptive structure described by
the transference y=f(x), where
x∈𝒳⊆ℝ denotes the system input random signal, having probability
density function px(x), and y∈𝒴⊆ℝ denotes the output
signal, having probability density function py(y), as shown in Figure 1. In the hypothesis that the neural system
transference is strictly monotonic, namely f′(x)>0, for all x∈𝒳, the relationship between the input distribution, the
output distribution, and the system transfer function is known to be
[9]
py(y)=px(x)f′(x)|x=f−1(y),x∈𝒳,
where f−1(⋅) denotes the
inverse of function f(⋅). Usually, (1)
is interpreted as an analysis formula, which allows
computing the output distribution when the input distribution and the system
transference function are known. However, the cardinal equation (1)
may also be interpreted as a formula that allows for
designing the nonlinear system when the distribution
px(⋅) is known and it
is desired that the system responds according to a
desired distribution py(⋅). In fact, (1)
may be rewritten as the differential equation:
f′(x)=px(x)py(f(x)),x∈𝒳.
In general, such design operation is rather difficult, because
(2) in the unknown
f(⋅) involves the solution of a nonlinear differential equation, provided that a consistent
boundary condition is specified.
Neural system, neural dual system, input/output sample spaces and
their statistical distributions.
In the recent contribution [10],
we presented a pseudorandom samples generator based on a nonlinear
monotonic neural system, whose transference function is denoted
by f(⋅), tuned on the basis of the differential equation
(2).
The cardinal design
equation (2) was
proposed to be solved via a (relaxation-type)
fixed-point algorithm. The key advantages of the method proposed in
[10] are as follows. (a) In order to obtain a
fully-tunable neural transference function, a look-up-table
representation was chosen. It guarantees high flexibility in
the shape of the neural transference as well as easiness of
representation and handling of the involved quantities. (b)
The fixed-point learning algorithm exhibits fast convergence over other
possible methods such as the gradient-based one: unlike these methods, the
fixed-point learning algorithm does not require the computation of derivatives
of the involved functions.
The resulting random-number generation method should
be thus read as a two-stage procedure. The first stage consists
in solving the cardinal differential equation
(2) in the unknown function
f(⋅), given the distributions
px(⋅) and py(⋅) as data. The second stage consists in
generating input random samples drawn from the
distribution px(⋅), then letting such random samples pass through the
learnt nonlinear neural system by computing output values y=f(x). The random samples y are assured to
be distributed according to the probability density function py(⋅).
However, we recognized that the method presented in [10] also suffers from some
drawbacks, namely the following. (a) For numerical
convergence purpose, each step of the fixed-point-type tuning
algorithm needed to be followed by some normalization
steps. Namely, from (2),
it is easily seen that when the function
py(f(x)) approaches 0, the computation of f′(x) becomes ill-conditioned,
therefore the quantity py(f(x)) was replaced by py(f(x))+γ, with γ>0 being a small
constant to be properly sized. Also, in order to refine learning, after each
iteration step, the solution f(x) needed to be
normalized either by affine scaling, in order to control the range of variable y , or by linear scaling in order to match the true
value of output distribution moment of preselected order. This, in turn, requires
computing in advance the (closed form) moments of interest of the output
distribution. (b) In spite of affine scaling, it was not easy to control the
range of the output value y, as affine scaling does not guarantee convergence in
every case of interest, therefore it could not be employed in every case. (c)
The developed procedure was customized to generate output distributions that
are either symmetric (namely, py(−y)=py(y)) or completely
skewed to the right (namely, py(y)=0, for all y<0) only.
Asymmetric or general-shape distributions were not considered.
In the present paper, we consider the problem of extending the
previous method to the generation of asymmetric distributions by
removing the constraint of symmetry or skewedness to the right.
Also, we propose a way to avoid probability
density function. The solution of choice implies a change in the viewpoint
of cardinal equation (1):
instead of converting formula (1) into
the differential equation
(2), we convert it into a
new differential equation, hereafter referred to as
dual cardinal equation, which will prove easier
to solve and more flexible to use in practice, while retaining the
previous numerical representation/advantages. Thus, we will
retain the effective numerical representation of the involved
quantity already introduced in the works
[10, 11], based on the “look-up table”
(LUT) implementation of neural activation functions as well as
the efficient numerical algorithm to solve the dual cardinal
equation. LUTs were proven to provide an efficient way of
representing and handling the variables appearing within the
devised random number generation algorithm. A prominent advantage
of the procedure is the lack of hard computational requirements
except for LUT handling, which consists of sorting/searching on
lists of numbers and of few simple algebraic operations on
numbers.
The effectiveness of the proposed approach will be
evaluated through numerical experiments. In particular, the designed
experiments followed a logical succession, beginning with a basic assessment of
the proposed method when applied to bi-Gaussian distribution,
which is then followed by comparably more difficult distributions,
namely a generalized Gaussian distribution and an asymmetric
Gamma distribution.
The existing method presented in [3]
is worth discussing. It concerns a neural-networks-type algorithm to generate random
vectors with arbitrary marginal distributions and correlation
matrix, based on NORTA method. The “normal-to-anything” (NORTA)
method (see, e.g., [12]) is
one of the most efficient methods for random vector generation.
In [3], a technique
was presented to generate the correlation matrix
of normal random vectors based on an artificial neural networks
approach. The NORTA algorithm works in the following way to generate
random samples with prescribed probability density function.
First, generate zero-mean unit-variance random samples
xi,
i∈{1,⋯,Q}. Then, generate the desired random samples as
yi=Py−1(Φ(xi)), where Φ(⋅) denotes the
cumulative distribution function of a standard normal random variable and Py(⋅) denotes the
desired cumulative distribution function, with Py−1(u)=inf{z∣Py(z)≥u}, u∈[01]. It appears, thus, as a transformation method.
Most of the methods of random vector generation known
from the literature impose constraints on the size of the random
vectors and many of them are applicable only for bivariate
distributions whose components are equidistributed.
Conversely, within the NORTA framework, marginal probability distributions
for vector components as well as their correlation matrix may be
specified. Obtaining the prescribed generated random vector
correlation matrix requires solving an involved nonlinear system
of equations, which is the most serious problem in this kind
of approach. Paper [3] makes use of a
multilayer perceptron neural network to estimate correlation
matrices of normal random vectors, allowing thus to
overcome the analytically involved equations of NORTA algorithm.
While the method proposed here is more general than NORTA in the
sense that it works for any kind of available generator
(not only Gaussian), it is less general in the sense that it
does not allow to generate multivariate random variables with
prescribed joint statistics.
2. Dual Cardinal Equation AND Its Numerical Solution
The present section formalizes the learning problem at hand
and illustrates a fixed-point-based numerical algorithm to
solve the dual cardinal equation.
2.1. Dual cardinal equation and neural system
The key point of the new method consists in learning the inverse function
f−1(⋅) instead of the
function f(⋅). As it will be clarified in the next sections, this
choice simplifies the learning problem while adding slight
computational burden to the usage of the learnt neural system as
a generative model.
We denote by x=g(y)=deff−1(y) the inverse
function of the actual neural transfer function and refer to the new neural
system, having g(⋅) as transfer
function, as the “dual neural system” (shown in Figure
1). The purpose here is to learn
a dual neural system that warps py(⋅) into px(⋅) under the
constraint g′(y)>0, for all y∈𝒴. We denote the interval of interest for the generated
random variable as 𝒴=[y̲y¯]. With this hypothesis on the
nonlinear dual neural transfer function, the cardinal equation (1)
may be rewritten as
g′(y)=py(y)px(g(y)),g(y̲)=0,y∈𝒴,
which will be hereafter referred to as “dual cardinal equation.”
It is worth noting that the boundary condition
g(y̲)=0 is completely arbitrary. While there are no
theoretical reasons to set the boundary condition in any
specific way, the above choice is motivated by the observation
that it simplifies the fixed-point adapting algorithm with
respect to the previous version proposed in
[10].
In general, a closed-form solution to
(3) may not be realized,
thus we should resort to an iterative learning algorithm to
search for a solution. Formally, this means designing an
algorithm that generates a succession of functions
gn(y), n∈ℕ, whose limit coincides to the solution of
(3). A way to generate
such a succession is to employ the fixed-point algorithm:
gn+1(y)=∫y̲ypy(t)dtpx(gn(t)),n≥0,y∈𝒴. As a figure-of-convergence of learning process,
we consider the weighted difference of function
g(⋅) between two successive iterations, namely,
Δgn=def∫𝒴|gn(y)−gn−1(y)|py(y)dy,n≥1. As initial guess, we assume
g0(y)=0, for all y∈𝒴.
After learning an inverse function g(⋅), the numerical procedure should calculate the actual
nonlinear function f(⋅) by numerical
inversion. As it will be clarified in the next section, within the framework
proposed here, such operation involves a very little
computational effort.
2.2. Numerical implementation of the learning procedure
From an implementation viewpoint, the algorithm
(4) needs to be
discretized in order to obtain a version suitable to be
implemented on a computer.
We choose to represent function
gn(y) by a numerical vector: in practice, we suppose the interval 𝒴=[y̲yy¯] of interest to
be partitioned into N≥1 discrete bins. This gives rise to the vector-type
representation y∈ℝN+1 of the support
of the output sequence probability density function, where
y contains N+1 values
regularly spaced in 𝒴 with
spacing-width denoted as Δy. Then gn(y) may be
represented by a numerical vector gn∈ℝN+1 and the neural
input-output transference is now represented by the discrete relationship (g,y)∈ℝN+1×ℝN+1, namely, a numerical look-up table. The
entries of a vector gn may be denoted
by an extra footer, that is, by gn,k, with k∈{0,1,…,N}. The interval Δy relates to the
integer N and may be defined
as Δy=def(y¯−y̲)/N.
In order to translate the learning rule
(4) into a version
suitable to numerical representation, we should consider the
inherent limitations of numerical integration of differential
equations. The following notes are worth taking into account.
(a) Output support selection: the ultimate
purpose of the random number generation method under construction
is to generate random samples with desired probability
distribution within a range of interest,
namely, with values within an interval that is deemed suitable
for the purposes that random samples generation is launched for.
Therefore, the output range
𝒴=[y̲yy¯] is to be freely
selected according to the needs the random samples are to be generated for.
Then, the above-mentioned vector y has entries yk computed as yk=y̲+k⋅Δy, with k∈{0,1,…,N}. (b) Input support selection: in
order to prevent the denominator of the quantity gn+1'(y) in (4)
to become too close to zero, a sensible choice is to
carefully select the support 𝒳. As in this paper we consider the input probability
density function to be either (symmetric) Gaussian or uniform, we set 𝒳=[−RxyRx], with Rx>0. The value of constant Rx is to be
selected in such a way that px(Rx)≫0. It is worth recalling that the support of the input
distribution may be arbitrarily selected as it does not affect the support of
the output distribution. (c) Iterative range scaling:
after each learning step, an affine normalization operation is
performed, that linearly scales the entries of the putative
solution gn so that gn,0=−Rx and gn,N=Rx.
In order to describe the numerical learning algorithm,
the following operators are defined for a generic look-up table (h,y)∈ℝN+1×ℝN+1:
cumsum(h)0=0,cumsum(h)k=def∑i=0k−1giΔy,affscale{h;a,b}k=defa+(hk−min{h})(b−a)max{h}−min{h}, where the subscript
k denotes the
kth entry of the
vectors cumsum(h) and affscale{h;a,b}. The behavior of the “cumsum” operator is
illustrated in Figure 2, which also provides a
visual representation of look-up tables. In practice, the
considered numerical version of the learning rule
(4) writes
(A0)g0:=0,(A1)g′n+1:=pypx(gn),n≥0,(A2)gn+1:=cumsum{g′n+1},(A3)gn+1:=affscale{gn+1;−Rx,Rx},
where symbol := denotes vector
values assignment and
py denotes the
vector of N+1 entries
containing the values of py(⋅) corresponding
to the values in y, and its entries may be denoted as
pyk, with
k∈{0,1,…,N}.
Behavior of the “cumsum” operator for look-up tables.
In terms of look-up-tables entries, the learning relaxation index
Δgn of definition
(5) may be approximated as
Δgn≈∑k=0N|gn,k−gn−1,k|pykΔy,n≥1.
2.3. Use of the neural system as
generative model
When a suitable dual neural system described by the transference g(⋅) has been
learnt, it may be effectively used to generate
random samples drawn from the desired statistical distribution.
The number of available input samples (that coincides with the
number of output samples to be generated) is hereafter
denoted by Q. The difficulty here is that the input samples x are known while the output samples
y are supposed to be computed as
y=f(x). However, unlike in [10], the function f(⋅) is not known in the present setting as its inverse g(⋅) only has been learnt. Nevertheless, the inversion
of function g(⋅) is not required in order to employ the dual
neural system as a generative model, provided an
appropriate usage of the look-up table representing g(⋅) is made. First, it is necessary to produce a
realization {xi}, i∈{1,…,Q}, drawn from the available-generator distribution px(⋅) (having, e.g.,
zero-mean Gaussian or uniform probability density function) ranging in 𝒳. About generation of input samples, as they are
generated by using an available generator whose range is wider than 𝒳, some generated input samples will be necessarily
discarded. The amount of discarded input samples may be quantified. Let us
denote by Px(⋅) the cumulative distribution function of the
input, namely,
Px(x)=def∫−∞xpx(t)dt.
The ratio ρ of the number of discarded samples over the
total number of generated samples is given by
ρ(Rx)=def#discardedsamples#generatedsamples=1−2Px(Rx).
The parameter Rx
may thus be selected in order to adjust the value of
ρ(Rx)
to design needs. Then, it is necessary to address
the proper values in the learnt look-up table (g,y)∈ℝN+1×ℝN+1corresponding to the values of {xi}, i∈{1,…,Q}, by finding pointers ri∈{0,1,…,N} such that gri≈xi. This means searching, in the whole look-up table, for the
closest value of g(⋅) to the sample xi. Such operation should be performed in an efficient way.
Finally, the desired set {y˜i} of output samples, approximately
distributed according to the probability density function py(⋅), may be obtained by setting y˜i:=yri,i∈{1,…,Q}, where
yri denotes the
rith entry of the look-up table
(g,y). (Commented MATLAB code is available on request.)
3. Computer-based Numerical Experiments
In the following experiments, we consider generating random univariate
samples with prescribed density function within prescribed ranges of interest,
supposing that a prototype Gaussian random number generator is available.
The prototype Gaussian distribution has zero mean and unitary variance.
The parameter Rx was set to 1 in all the experiments, which corresponds to a ratio ρ≈0.3173 that allows retaining
about 70% of the generated input samples. The experiments were run on a
1.86 GHz, 512 MB-RAM
platform.
3.1. Experiments on a “bi-Gaussian” distribution
The first case of generation of a random variable concerns a
“bi-Gaussian” distribution defined by
G2(y)=12[12πσ1exp(−(y−μ1)22σ12)+12πσ2exp(−(y−μ2)22σ22)] that may assume fairly asymmetric shapes.
The numerical results presented below pertain to values σ1=0.3, μ1=−0.5, σ2=1 and
μ2=0.8.
The interval of interest for the output variable is set to
𝒴=[−2.54].
The total number of generated output samples amounts to
Q=68219. The number of points in which the function g(⋅) is computed is N=1000.
The results obtained by running the learning algorithm
(7) are shown in Figure
3. The values of the index Δgn shows that the
fixed-point algorithm may be stopped after 5 iterations. In Figure
3, the histogram estimates (with 50 bins) of the
generated Gaussian data and of the bi-Gaussian output–-obtained with the
learnt dual system–-may be observed as well.
Result of dual neural system adaptation with Gaussian input and
bi-Gaussian output.
Cumulative results on repeated independent trials are
illustrated. The number of iterations of the algorithm
(7) was set to 10, while the other data
stayed the same of the previous single-run experiment. The number N of points in which the
function g(⋅) was computed ranged
from 200 to 2000 with step 200, in order to obtain some information about the sensitivity
of the algorithm to the selection of the number of points in the domain 𝒴 and about the
influence of the number N in the computational
complexity of the algorithm. In particular, the sensitivity of the algorithm
with respect to the number N was measured via a
discrepancy index DSC computed as follows. (a) The histogram-based estimate of
the probability density function of the generated samples is computed on a
number of bins equal to 50. The discrete values of such estimate are denoted by p^yb, b∈{1,2,…,50}. (b) The true values of the probability density function py(⋅) are computed in
correspondence of the histogram's bin-centers. The discrete values of such
probability density function are denoted by pyb, b∈{1,2,…,50}. (c) The weighted-square-difference-type discrepancy index
is computed by the expression
DSC=def∑b=150(p^yb−pyb)2pyb.
The average number of generated samples varies between
about 68250 and 68290. The obtained results are summarized in
Tables and . The tables show the average run-time required for
learning (expressed in seconds), the average run-time required to
generate the samples (use of the learnt systems as a generative model) and average DSC index
value. As it is readily appreciated, the computational complexity owing to the
learning phase depends on the number of points used to approximate the
nonlinear transference g(⋅) as expected, while the computational
complexity owing to the generation phase depends only
slightly on N. The sensitivity of the method measured by the discrepancy
index DSC is high for low values of the parameter
N, while it
becomes quite low for values of
N larger than 1000.
Average results about the experiment on bi-Gaussian
random number generation; averages computed over 100
independent trials when the algorithm
(7) was iterated 10 times (first batch of results).
POINTS N
200
400
600
800
1000
AVG. LEARN. TIME
0.0092
0.0090
0.0109
0.0117
0.0133
AVG. GEN. TIME
0.0517
0.0514
0.0509
0.0508
0.0509
AVG. DSC
0.0026
0.0018
0.0011
0.0009
0.0008
Average results about the experiment on bi-Gaussian
random number generation; averages computed over 100
independent trials when the
algorithm (7)
was iterated 10 times (second batch of results).
POINTS N
1200
1400
1600
1800
2000
AVG. LEARN. TIME
0.0145
0.0155
0.0176
0.0197
0.0209
AVG. GEN. TIME
0.0527
0.0528
0.0527
0.0511
0.0523
AVG. DSC
0.0007
0.0006
0.0005
0.0006
0.0005
3.2. Experiments on a generalized Gaussian
distribution
The second example of random samples generation is about a
generalized Gaussian distribution
[13]:
T(y)=sechλ−1(y−μ)−λsechλ+1(y−μ)sinh2(y−μ)2πσ×exp[−(sinh(y)sechλ(y)−sinh(μ)sechλ(μ))22σ2],
where sinh(⋅) denotes the hyperbolic sine function and sech(⋅) denotes the reciprocal
of the hyperbolic cosine function (namely, the hyperbolic secant function) The
present generalized Gaussian distribution (GGD) differs by the standard GGD
model encountered in literature
(see, e.g., [14]). It belongs to the
general exponential family of distributions of the type py(y)∝exp(−κ2(y)), with κ(⋅) satisfying appropriate
compatibility conditions. The distribution
(12) as well as the GGD
in [14] belong to the above
exponential family.
The numerical results presented below pertain to
values σ=1, μ=0.8, and λ=0.5. The interval of interest for the output variable is set to 𝒴=[−34]. The total number of generated output samples amounts to Q=68335. The number of points in which the function g(⋅) is computed is N=1000.
The results obtained by running the learning algorithm
(7) are shown in Figure
4. The values of the index Δgn show that the fixed-point algorithm may be
safely stopped after 5 iterations again. In Figure
4, the histogram estimates
(with 50 bins) of the generated Gaussian data and of
the generalized Gaussian output may be observed as well.
Result of dual neural
system adaptation with Gaussian input and
generalized Gaussian output.
Cumulative results are illustrated as well. The number
of iterations of the algorithm
(7) was set to 10, while
the other data stayed the same of the previous single-run
experiment. The number Nof points ranged from 200 to 1000 with step 200. The average number of generated samples varies between
about 68250 and 68290. The obtained results are summarized in Table
.
Average results about the experiment on generalized
Gaussian random number generation; averages computed
over 100 independent trials when the algorithm
(7) was
iterated 10 times.
POINTS N
200
400
600
800
1000
AVG. LEARN. TIME
0.0129
0.0159
0.0229
0.0281
0.0320
AVG. GEN. TIME
0.0511
0.0517
0.0500
0.0502
0.0513
AVG. DSC
0.0038
0.0018
0.0011
0.0007
0.0005
3.3. Experiments on a Gamma distribution
The third example is repeated from
[10]: we considered the generation of a (symmetric) Gamma
distribution:
B(y)=defαβ1/α2Γ(1/α)exp(−β|y|α).
This choice is motivated by the observation that the random number generation algorithm
in [10] gives rise to the
most inaccurate result when tested on the Gamma distribution .
The numerical results presented below pertain to
values α=0.8 and β=4.
The interval of interest for the output variable is set to
𝒴=[−22]. The total number of
generated output samples amounts to
Q=68355. The number of points in which the
function g(⋅) is computed is N=1500. The results obtained by running the
learning algorithm (7)
are shown in Figure 5. The values
of the index Δgn show that the fixed-point algorithm may be safely
stopped after 5 iterations again. Figure
5 shows the histogram estimates
(with 50 bins) of the generated Gaussian data and of the
Gamma-distributed output.
Result of dual neural
system adaptation with Gaussian input and
Gamma output.
Cumulative results were obtained by setting the number
of iterations of the algorithm
(7) to 20, while the
other data stayed the same of the previous single-run experiment.
The number Nof points ranged from 1000 to 1800 with step 200. The average number of generated samples varies between
about 68230 and 68280. The obtained results are summarized in Table
.
Average results about the experiment on Gamma random
number generation; averages computed over 100 independent
trials when the algorithm
(7)
was iterated 20 times.
POINTS N
1000
1200
1400
1600
1800
AVG. LEARN. TIME
0.0165
0.0176
0.0220
0.0242
0.0261
AVG. GEN. TIME
0.0516
0.0516
0.0504
0.0514
0.0513
AVG. DSC
0.0137
0.0118
0.0118
0.0109
0.0101
4. Conclusion
The aim of the present manuscript was to present a novel
random number generation technique based on dual neural
system learning. We elaborated over our recent work
[10] in order to
obtain a new learning algorithm free of the need of choosing
parameters and normalization-criteria. The main idea is to shift
the learning paradigm from the viewpoint of cardinal equation
solving to dual cardinal equation solving, which appears to be
more easily profitable.
The proposed numerical results confirmed the agreement
between the desired and obtained distributions of the generated
variate. The analysis of computational burden, in terms of
running times, shows that the proposed algorithm is not
computationally demanding.
L'EcuyerP.BanksJ.Random number generation1998New York, NY, USAJohn Wiley & Sons93137chapter 4LagariasJ. C.PomeranceC.Pseudorandom number generators in cryptography and number theory199042Providence, RI, USAAmerican Mathematical Society115143NiakiS. T. A.AbbasiB.NORTA and neural networks based method to generate RANDOM vectors with arbitrary marginal distributions and correlation matrixProceedings of the 17th IASTED International Conference on Modelling and SimulationMay 2006Montreal, Canada234239MarsagliaG.BillardL.A current view of random number generators1985Amsterdam, The NetherlandsElsevier310NiederreiterH.1992Philadelphia, Pa, USASIAMRipleyB. D.Thoughts on pseudorandom number generators199031115316310.1016/0377-0427(90)90346-2CichockiA.AmariS.-I.2002New York, NY, USAJohn Wiley & SonsKnuthD. E.199723rdReading, Mass, USAAddison-WesleyPapoulisA.1996Englewood Cliffs, NJ, USAPrentice-HallFioriS.Neural systems with numerically-matched input-output statistic: variate generation200623214317010.1007/s11063-005-4016-6FioriS.Neural systems with numerically matched input-output statistic: isotonic bivariate statistical modeling20072007237185910.1155/2007/71859GhoshS.HendersonS. G.YücesanE.ChenC.-H.SnowdonJ. L.CharnesJ. M.Properties of the NORTA method in higher dimensions1Proceedings of the Winter Simulation Conference (WSC '02)December 2002San Diego, Calif, USA26326910.1109/WSC.2002.1172894TsaiA. C.2006JanuaryPersonal communicationKokkinakisK.NandiA. K.Exponent parameter estimation for generalized Gaussian probability density functions with application to speech modeling20058591852185810.1016/j.sigpro.2005.02.017