Abstract: In this paper, a new probability density function (pdf)
is proposed to model the statistics of wavelet coefficients, and a
simple Kalman-s filter is derived from the new pdf using Bayesian
estimation theory. Specifically, we decompose the speckled image
into wavelet subbands, we apply the Kalman-s filter to the high
subbands, and reconstruct a despeckled image from the modified
detail coefficients. Experimental results demonstrate that our method
compares favorably to several other despeckling methods on test
synthetic aperture radar (SAR) images.
Abstract: This paper discusses EM algorithm and Bootstrap
approach combination applied for the improvement of the satellite
image fusion process. This novel satellite image fusion method based
on estimation theory EM algorithm and reinforced by Bootstrap
approach was successfully implemented and tested. The sensor
images are firstly split by a Bayesian segmentation method to
determine a joint region map for the fused image. Then, we use the
EM algorithm in conjunction with the Bootstrap approach to develop
the bootstrap EM fusion algorithm, hence producing the fused
targeted image. We proposed in this research to estimate the
statistical parameters from some iterative equations of the EM
algorithm relying on a reference of representative Bootstrap samples
of images. Sizes of those samples are determined from a new
criterion called 'hybrid criterion'. Consequently, the obtained results
of our work show that using the Bootstrap EM (BEM) in image
fusion improve performances of estimated parameters which involve
amelioration of the fused image quality; and reduce the computing
time during the fusion process.
Abstract: The Bayesian Optimization Algorithm (BOA) is an algorithm based on the estimation of distributions. It uses techniques from modeling data by Bayesian networks to estimating the joint distribution of promising solutions. To obtain the structure of Bayesian network, different search algorithms can be used. The key point that BOA addresses is whether the constructed Bayesian network could generate new and useful solutions (strings), which could lead the algorithm in the right direction to solve the problem. Undoubtedly, this ability is a crucial factor of the efficiency of BOA. Varied search algorithms can be used in BOA, but their performances are different. For choosing better ones, certain suitable method to present their ability difference is needed. In this paper, a greedy search algorithm and a stochastic search algorithm are used in BOA to solve certain optimization problem. A method using Kullback-Leibler (KL) Divergence to reflect their difference is described.
Abstract: Trust management and Reputation models are
becoming integral part of Internet based applications such as CSCW,
E-commerce and Grid Computing. Also the trust dimension is a
significant social structure and key to social relations within a
collaborative community. Collaborative Decision Making (CDM) is
a difficult task in the context of distributed environment (information
across different geographical locations) and multidisciplinary
decisions are involved such as Virtual Organization (VO). To aid
team decision making in VO, Decision Support System and social
network analysis approaches are integrated. In such situations social
learning helps an organization in terms of relationship, team
formation, partner selection etc. In this paper we focus on trust
learning. Trust learning is an important activity in terms of
information exchange, negotiation, collaboration and trust
assessment for cooperation among virtual team members. In this
paper we have proposed a reinforcement learning which enhances the
trust decision making capability of interacting agents during
collaboration in problem solving activity. Trust computational model
with learning that we present is adapted for best alternate selection of
new project in the organization. We verify our model in a multi-agent
simulation where the agents in the community learn to identify
trustworthy members, inconsistent behavior and conflicting behavior
of agents.
Abstract: In this contribution a newly developed elearning environment is presented, which incorporates Intelligent Agents and Computational Intelligence Techniques. The new e-learning environment is constituted by three parts, the E-learning platform Front-End, the Student Questioner Reasoning and the Student Model Agent. These parts are distributed geographically in dispersed computer servers, with main focus on the design and development of these subsystems through the use of new and emerging technologies. These parts are interconnected in an interoperable way, using web services for the integration of the subsystems, in order to enhance the user modelling procedure and achieve the goals of the learning process.
Abstract: In unsupervised segmentation context, we propose a bi-dimensional hidden Markov chain model (X,Y) that we adapt to the image segmentation problem. The bi-dimensional observed process Y = (Y 1, Y 2) is such that Y 1 represents the noisy image and Y 2 represents a noisy supplementary information on the image, for example a noisy proportion of pixels of the same type in a neighborhood of the current pixel. The proposed model can be seen as a competitive alternative to the Hilbert-Peano scan. We propose a bayesian algorithm to estimate parameters of the considered model. The performance of this algorithm is globally favorable, compared to the bi-dimensional EM algorithm through numerical and visual data.
Abstract: In this paper, we present a new learning algorithm for
anomaly based network intrusion detection using improved self
adaptive naïve Bayesian tree (NBTree), which induces a hybrid of
decision tree and naïve Bayesian classifier. The proposed approach
scales up the balance detections for different attack types and keeps
the false positives at acceptable level in intrusion detection. In
complex and dynamic large intrusion detection dataset, the detection
accuracy of naïve Bayesian classifier does not scale up as well as
decision tree. It has been successfully tested in other problem
domains that naïve Bayesian tree improves the classification rates in
large dataset. In naïve Bayesian tree nodes contain and split as
regular decision-trees, but the leaves contain naïve Bayesian
classifiers. The experimental results on KDD99 benchmark network
intrusion detection dataset demonstrate that this new approach scales
up the detection rates for different attack types and reduces false
positives in network intrusion detection.
Abstract: Numerous divergence measures (spectral distance, cepstral
distance, difference of the cepstral coefficients, Kullback-Leibler
divergence, distance given by the General Likelihood Ratio, distance
defined by the Recursive Bayesian Changepoint Detector and the
Mahalanobis measure) are compared in this study. The measures are
used for detection of abrupt spectral changes in synthetic AR signals
via the sliding window algorithm. Two experiments are performed;
the first is focused on detection of single boundary while the second
concentrates on detection of a couple of boundaries. Accuracy of
detection is judged for each method; the measures are compared
according to results of both experiments.
Abstract: The paper presents an applied study of a multivariate AR(p) process fitted to daily data from U.S. commodity futures markets with the use of Bayesian statistics. In the first part a detailed description of the methods used is given. In the second part two BVAR models are chosen one with assumption of lognormal, the second with normal distribution of prices conditioned on the parameters. For a comparison two simple benchmark models are chosen that are commonly used in todays Financial Mathematics. The article compares the quality of predictions of all the models, tries to find an adequate rate of forgetting of information and questions the validity of Efficient Market Hypothesis in the semi-strong form.
Abstract: In this paper, we propose a texture feature-based
language identification using wavelet-domain BDIP (block difference
of inverse probabilities) and BVLC (block variance of local
correlation coefficients) features and FFT (fast Fourier transform)
feature. In the proposed method, wavelet subbands are first obtained
by wavelet transform from a test image and denoised by Donoho-s
soft-thresholding. BDIP and BVLC operators are next applied to the
wavelet subbands. FFT blocks are also obtained by 2D (twodimensional)
FFT from the blocks into which the test image is
partitioned. Some significant FFT coefficients in each block are
selected and magnitude operator is applied to them. Moments for each
subband of BDIP and BVLC and for each magnitude of significant
FFT coefficients are then computed and fused into a feature vector. In
classification, a stabilized Bayesian classifier, which adopts variance
thresholding, searches the training feature vector most similar to the
test feature vector. Experimental results show that the proposed
method with the three operations yields excellent language
identification even with rather low feature dimension.
Abstract: Nowadays predicting political risk level of country
has become a critical issue for investors who intend to achieve
accurate information concerning stability of the business
environments. Since, most of the times investors are layman and
nonprofessional IT personnel; this paper aims to propose a
framework named GECR in order to help nonexpert persons to
discover political risk stability across time based on the political
news and events.
To achieve this goal, the Bayesian Networks approach was
utilized for 186 political news of Pakistan as sample dataset.
Bayesian Networks as an artificial intelligence approach has been
employed in presented framework, since this is a powerful technique
that can be applied to model uncertain domains. The results showed
that our framework along with Bayesian Networks as decision
support tool, predicted the political risk level with a high degree of
accuracy.
Abstract: Data mining incorporates a group of statistical
methods used to analyze a set of information, or a data set. It operates
with models and algorithms, which are powerful tools with the great
potential. They can help people to understand the patterns in certain
chunk of information so it is obvious that the data mining tools have
a wide area of applications. For example in the theoretical chemistry
data mining tools can be used to predict moleculeproperties or
improve computer-assisted drug design. Classification analysis is one
of the major data mining methodologies. The aim of thecontribution
is to create a classification model, which would be able to deal with a
huge data set with high accuracy. For this purpose logistic regression,
Bayesian logistic regression and random forest models were built
using R software. TheBayesian logistic regression in Latent GOLD
software was created as well. These classification methods belong to
supervised learning methods.
It was necessary to reduce data matrix dimension before construct
models and thus the factor analysis (FA) was used. Those models
were applied to predict the biological activity of molecules, potential
new drug candidates.
Abstract: This work presents a fusion of Log Gabor Wavelet
(LGW) and Maximum a Posteriori (MAP) estimator as a speech
enhancement tool for acoustical background noise reduction. The
probability density function (pdf) of the speech spectral amplitude is
approximated by a Generalized Laplacian Distribution (GLD).
Compared to earlier estimators the proposed method estimates the
underlying statistical model more accurately by appropriately
choosing the model parameters of GLD. Experimental results show
that the proposed estimator yields a higher improvement in
Segmental Signal-to-Noise Ratio (S-SNR) and lower Log-Spectral
Distortion (LSD) in two different noisy environments compared to
other estimators.
Abstract: Testing accounts for the major percentage of technical
contribution in the software development process. Typically, it
consumes more than 50 percent of the total cost of developing a
piece of software. The selection of software tests is a very important
activity within this process to ensure the software reliability
requirements are met. Generally tests are run to achieve maximum
coverage of the software code and very little attention is given to the
achieved reliability of the software. Using an existing methodology,
this paper describes how to use Bayesian Belief Networks (BBNs) to
select unit tests based on their contribution to the reliability of the
module under consideration. In particular the work examines how the
approach can enhance test-first development by assessing the quality
of test suites resulting from this development methodology and
providing insight into additional tests that can significantly reduce
the achieved reliability. In this way the method can produce an
optimal selection of inputs and the order in which the tests are
executed to maximize the software reliability. To illustrate this
approach, a belief network is constructed for a modern software
system incorporating the expert opinion, expressed through
probabilities of the relative quality of the elements of the software,
and the potential effectiveness of the software tests. The steps
involved in constructing the Bayesian Network are explained as is a
method to allow for the test suite resulting from test-driven
development.
Abstract: Software developed for a specific customer under contract
typically undergoes a period of testing by the customer before
acceptance. This is known as user acceptance testing and the process
can reveal both defects in the system and requests for changes to
the product. This paper uses nonhomogeneous Poisson processes to
model a real user acceptance data set from a recently developed
system. In particular a split Poisson process is shown to provide an
excellent fit to the data. The paper explains how this model can be
used to aid the allocation of resources through the accurate prediction
of occurrences both during the acceptance testing phase and before
this activity begins.
Abstract: Dengue disease is an infectious vector-borne viral
disease that is commonly found in tropical and sub-tropical regions,
especially in urban and semi-urban areas, around the world and
including Malaysia. There is no currently available vaccine or
chemotherapy for the prevention or treatment of dengue disease.
Therefore prevention and treatment of the disease depend on vector
surveillance and control measures. Disease risk mapping has been
recognized as an important tool in the prevention and control
strategies for diseases. The choice of statistical model used for
relative risk estimation is important as a good model will
subsequently produce a good disease risk map. Therefore, the aim of
this study is to estimate the relative risk for dengue disease based
initially on the most common statistic used in disease mapping called
Standardized Morbidity Ratio (SMR) and one of the earliest
applications of Bayesian methodology called Poisson-gamma model.
This paper begins by providing a review of the SMR method, which
we then apply to dengue data of Perak, Malaysia. We then fit an
extension of the SMR method, which is the Poisson-gamma model.
Both results are displayed and compared using graph, tables and
maps. Results of the analysis shows that the latter method gives a
better relative risk estimates compared with using the SMR. The
Poisson-gamma model has been demonstrated can overcome the
problem of SMR when there is no observed dengue cases in certain
regions. However, covariate adjustment in this model is difficult and
there is no possibility for allowing spatial correlation between risks in
adjacent areas. The drawbacks of this model have motivated many
researchers to propose other alternative methods for estimating the
risk.
Abstract: There are three approaches to complete Bayesian
Network (BN) model construction: total expert-centred, total datacentred,
and semi data-centred. These three approaches constitute the
basis of the empirical investigation undertaken and reported in this
paper. The objective is to determine, amongst these three
approaches, which is the optimal approach for the construction of a
BN-based model for the performance assessment of students-
laboratory work in a virtual electronic laboratory environment. BN
models were constructed using all three approaches, with respect to
the focus domain, and compared using a set of optimality criteria. In
addition, the impact of the size and source of the training, on the
performance of total data-centred and semi data-centred models was
investigated. The results of the investigation provide additional
insight for BN model constructors and contribute to literature
providing supportive evidence for the conceptual feasibility and
efficiency of structure and parameter learning from data. In addition,
the results highlight other interesting themes.
Abstract: This paper applies Bayesian Networks to support
information extraction from unstructured, ungrammatical, and
incoherent data sources for semantic annotation. A tool has been
developed that combines ontologies, machine learning, and
information extraction and probabilistic reasoning techniques to
support the extraction process. Data acquisition is performed with the
aid of knowledge specified in the form of ontology. Due to the
variable size of information available on different data sources, it is
often the case that the extracted data contains missing values for
certain variables of interest. It is desirable in such situations to
predict the missing values. The methodology, presented in this paper,
first learns a Bayesian network from the training data and then uses it
to predict missing data and to resolve conflicts. Experiments have
been conducted to analyze the performance of the presented
methodology. The results look promising as the methodology
achieves high degree of precision and recall for information
extraction and reasonably good accuracy for predicting missing
values.
Abstract: It is a one-sided hypothesis testing process for assessing bioequivalence. Bootstrap and modified large-sample(MLS) methods are considered to study individual bioequivalence(IBE), type I error and power of hypothesis tests are simulated and compared with FDA(2001). The results show that modified large-sample method is equivalent to the method of FDA(2001) .
Abstract: Studies in neuroscience suggest that both global and
local feature information are crucial for perception and recognition of
faces. It is widely believed that local feature is less sensitive to
variations caused by illumination, expression and illumination. In
this paper, we target at designing and learning local features for face
recognition. We designed three types of local features. They are
semi-global feature, local patch feature and tangent shape feature.
The designing of semi-global feature aims at taking advantage of
global-like feature and meanwhile avoiding suppressing AdaBoost
algorithm in boosting weak classifies established from small local
patches. The designing of local patch feature targets at automatically
selecting discriminative features, and is thus different with traditional
ways, in which local patches are usually selected manually to cover
the salient facial components. Also, shape feature is considered in
this paper for frontal view face recognition. These features are
selected and combined under the framework of boosting algorithm
and cascade structure. The experimental results demonstrate that the
proposed approach outperforms the standard eigenface method and
Bayesian method. Moreover, the selected local features and
observations in the experiments are enlightening to researches in
local feature design in face recognition.