Abstract: Traditional principal components analysis (PCA)
techniques for face recognition are based on batch-mode training
using a pre-available image set. Real world applications require that
the training set be dynamic of evolving nature where within the
framework of continuous learning, new training images are
continuously added to the original set; this would trigger a costly
continuous re-computation of the eigen space representation via
repeating an entire batch-based training that includes the old and new
images. Incremental PCA methods allow adding new images and
updating the PCA representation. In this paper, two incremental
PCA approaches, CCIPCA and IPCA, are examined and compared.
Besides, different learning and testing strategies are proposed and
applied to the two algorithms. The results suggest that batch PCA is
inferior to both incremental approaches, and that all CCIPCAs are
practically equivalent.
Abstract: Discrimination between different classes of environmental
sounds is the goal of our work. The use of a sound recognition
system can offer concrete potentialities for surveillance and
security applications. The first paper contribution to this research
field is represented by a thorough investigation of the applicability
of state-of-the-art audio features in the domain of environmental
sound recognition. Additionally, a set of novel features obtained by
combining the basic parameters is introduced. The quality of the
features investigated is evaluated by a HMM-based classifier to which
a great interest was done. In fact, we propose to use a Multi-Style
training system based on HMMs: one recognizer is trained on a
database including different levels of background noises and is used
as a universal recognizer for every environment. In order to enhance
the system robustness by reducing the environmental variability, we
explore different adaptation algorithms including Maximum Likelihood
Linear Regression (MLLR), Maximum A Posteriori (MAP)
and the MAP/MLLR algorithm that combines MAP and MLLR.
Experimental evaluation shows that a rather good recognition rate
can be reached, even under important noise degradation conditions
when the system is fed by the convenient set of features.
Abstract: This paper presents a new technique for detection of
human faces within color images. The approach relies on image
segmentation based on skin color, features extracted from the two-dimensional
discrete cosine transform (DCT), and self-organizing
maps (SOM). After candidate skin regions are extracted, feature
vectors are constructed using DCT coefficients computed from those
regions. A supervised SOM training session is used to cluster feature
vectors into groups, and to assign “face" or “non-face" labels to those
clusters. Evaluation was performed using a new image database of
286 images, containing 1027 faces. After training, our detection
technique achieved a detection rate of 77.94% during subsequent
tests, with a false positive rate of 5.14%. To our knowledge, the
proposed technique is the first to combine DCT-based feature
extraction with a SOM for detecting human faces within color
images. It is also one of a few attempts to combine a feature-invariant
approach, such as color-based skin segmentation, together with
appearance-based face detection. The main advantage of the new
technique is its low computational requirements, in terms of both
processing speed and memory utilization.
Abstract: In wireless communication system, a Decision Feedback Equalizer (DFE) to cancel the intersymbol interference (ISI) is required. In this paper, an exact convergence analysis of the (DFE) adapted by the Least Mean Square (LMS) algorithm during the training phase is derived by taking into account the finite alphabet context of data transmission. This allows us to determine the shortest training sequence that allows to reach a given Mean Square Error (MSE). With the intention of avoiding the problem of ill-convergence, the paper proposes an initialization strategy for the blind decision directed (DD) algorithm. This then yields a semi-blind DFE with high speed and good convergence.
Abstract: This paper presents the prediction of kidney
dysfunction using different neural network (NN) approaches. Self
organization Maps (SOM), Probabilistic Neural Network (PNN) and
Multi Layer Perceptron Neural Network (MLPNN) trained with Back
Propagation Algorithm (BPA) are used in this study. Six hundred and
sixty three sets of analytical laboratory tests have been collected from
one of the private clinical laboratories in Baghdad. For each subject,
Serum urea and Serum creatinin levels have been analyzed and tested
by using clinical laboratory measurements. The collected urea and
cretinine levels are then used as inputs to the three NN models in
which the training process is done by different neural approaches.
SOM which is a class of unsupervised network whereas PNN and
BPNN are considered as class of supervised networks. These
networks are used as a classifier to predict whether kidney is normal
or it will have a dysfunction. The accuracy of prediction, sensitivity
and specificity were found for each type of the proposed networks
.We conclude that PNN gives faster and more accurate prediction of
kidney dysfunction and it works as promising tool for predicting of
routine kidney dysfunction from the clinical laboratory data.
Abstract: This paper explores the scalability issues associated
with solving the Named Entity Recognition (NER) problem using
Support Vector Machines (SVM) and high-dimensional features. The
performance results of a set of experiments conducted using binary
and multi-class SVM with increasing training data sizes are
examined. The NER domain chosen for these experiments is the
biomedical publications domain, especially selected due to its
importance and inherent challenges. A simple machine learning
approach is used that eliminates prior language knowledge such as
part-of-speech or noun phrase tagging thereby allowing for its
applicability across languages. No domain-specific knowledge is
included. The accuracy measures achieved are comparable to those
obtained using more complex approaches, which constitutes a
motivation to investigate ways to improve the scalability of multiclass
SVM in order to make the solution more practical and useable.
Improving training time of multi-class SVM would make support
vector machines a more viable and practical machine learning
solution for real-world problems with large datasets. An initial
prototype results in great improvement of the training time at the
expense of memory requirements.
Abstract: The purpose of this study is to investigate the
relationship between corporate governance and pricing for initial
public offerings (IPOs). Empirical result finds that the prediction of
pricing of IPOs with corporate governance added can have a rather
higher degree of predicting accuracy than that of non governance
added during the training and testing samples. Therefore, it can be
observed that corporate governance mechanism can affect the pricing
of IPOs
Abstract: Performance of any continuous speech recognition system is highly dependent on performance of the acoustic models. Generally, development of the robust spoken language technology relies on the availability of large amounts of data. Common way to cope with little data for training each state of Markov models is treebased state tying. This tying method applies contextual questions to tie states. Manual procedure for question generation suffers from human errors and is time consuming. Various automatically generated questions are used to construct decision tree. There are three approaches to generate questions to construct HMMs based on decision tree. One approach is based on misrecognized phonemes, another approach basically uses feature table and the other is based on state distributions corresponding to context-independent subword units. In this paper, all these methods of automatic question generation are applied to the decision tree on FARSDAT corpus in Persian language and their results are compared with those of manually generated questions. The results show that automatically generated questions yield much better results and can replace manually generated questions in Persian language.
Abstract: As a popular rank-reduced vector space approach,
Latent Semantic Indexing (LSI) has been used in information
retrieval and other applications. In this paper, an LSI-based content
vector model for text classification is presented, which constructs
multiple augmented category LSI spaces and classifies text by their
content. The model integrates the class discriminative information
from the training data and is equipped with several pertinent feature
selection and text classification algorithms. The proposed classifier
has been applied to email classification and its experiments on a
benchmark spam testing corpus (PU1) have shown that the approach
represents a competitive alternative to other email classifiers based
on the well-known SVM and naïve Bayes algorithms.
Abstract: Poly-β-hydroxybutyrate (PHB) is one of the most
famous biopolymers that has various applications in production of
biodegradable carriers. The most important strategy for enhancing
efficiency in production process and reducing the price of PHB, is the
accurate expression of kinetic model of products formation and
parameters that are effective on it, such as Dry Cell Weight (DCW)
and substrate consumption. Considering the high capabilities of
artificial neural networks in modeling and simulation of non-linear
systems such as biological and chemical industries that mainly are
multivariable systems, kinetic modeling of microbial production of
PHB that is a complex and non-linear biological process, the three
layers perceptron neural network model was used in this study.
Artificial neural network educates itself and finds the hidden laws
behind the data with mapping based on experimental data, of dry cell
weight, substrate concentration as input and PHB concentration as
output. For training the network, a series of experimental data for
PHB production from Hydrogenophaga Pseudoflava by glucose
carbon source was used. After training the network, two other
experimental data sets that have not intervened in the network
education, including dry cell concentration and substrate
concentration were applied as inputs to the network, and PHB
concentration was predicted by the network. Comparison of predicted
data by network and experimental data, indicated a high precision
predicted for both fructose and whey carbon sources. Also in present
study for better understanding of the ability of neural network in
modeling of biological processes, microbial production kinetic of
PHB by Leudeking-Piret experimental equation was modeled. The
Observed result indicated an accurate prediction of PHB
concentration by artificial neural network higher than Leudeking-
Piret model.
Abstract: In this paper, a comparative study of application of
supervised and unsupervised learning algorithms on illumination
invariant face recognition has been carried out. The supervised
learning has been carried out with the help of using a bi-layered
artificial neural network having one input, two hidden and one output
layer. The gradient descent with momentum and adaptive learning
rate back propagation learning algorithm has been used to implement
the supervised learning in a way that both the inputs and
corresponding outputs are provided at the time of training the
network, thus here is an inherent clustering and optimized learning of
weights which provide us with efficient results.. The unsupervised
learning has been implemented with the help of a modified
Counterpropagation network. The Counterpropagation network
involves the process of clustering followed by application of Outstar
rule to obtain the recognized face. The face recognition system has
been developed for recognizing faces which have varying
illumination intensities, where the database images vary in lighting
with respect to angle of illumination with horizontal and vertical
planes. The supervised and unsupervised learning algorithms have
been implemented and have been tested exhaustively, with and
without application of histogram equalization to get efficient results.
Abstract: Real world Speaker Identification (SI) application
differs from ideal or laboratory conditions causing perturbations that
leads to a mismatch between the training and testing environment
and degrade the performance drastically. Many strategies have been
adopted to cope with acoustical degradation; wavelet based Bayesian
marginal model is one of them. But Bayesian marginal models
cannot model the inter-scale statistical dependencies of different
wavelet scales. Simple nonlinear estimators for wavelet based
denoising assume that the wavelet coefficients in different scales are
independent in nature. However wavelet coefficients have significant
inter-scale dependency. This paper enhances this inter-scale
dependency property by a Circularly Symmetric Probability Density
Function (CS-PDF) related to the family of Spherically Invariant
Random Processes (SIRPs) in Log Gabor Wavelet (LGW) domain
and corresponding joint shrinkage estimator is derived by Maximum
a Posteriori (MAP) estimator. A framework is proposed based on
these to denoise speech signal for automatic speaker identification
problems. The robustness of the proposed framework is tested for
Text Independent Speaker Identification application on 100 speakers
of POLYCOST and 100 speakers of YOHO speech database in three
different noise environments. Experimental results show that the
proposed estimator yields a higher improvement in identification
accuracy compared to other estimators on popular Gaussian Mixture
Model (GMM) based speaker model and Mel-Frequency Cepstral
Coefficient (MFCC) features.
Abstract: Various models have been derived by studying large number of completed software projects from various organizations and applications to explore how project sizes mapped into project effort. But, still there is a need to prediction accuracy of the models. As Neuro-fuzzy based system is able to approximate the non-linear function with more precision. So, Neuro-Fuzzy system is used as a soft computing approach to generate model by formulating the relationship based on its training. In this paper, Neuro-Fuzzy technique is used for software estimation modeling of on NASA software project data and performance of the developed models are compared with the Halstead, Walston-Felix, Bailey-Basili and Doty Models mentioned in the literature.
Abstract: This article presents the evolution and technological changes implemented on the full scale simulators developed by the Simulation Department of the Instituto de Investigaciones Eléctricas1 (Mexican Electric Research Institute) and located at different training centers around the Mexican territory, and allows US to know the last updates, basically from the input/output view point, of the current simulators at some facilities of the electrical sector as well as the compatible industry of the electrical manufactures and industries such as Comision Federal de Electricidad (CFE*, The utility Mexican company). Tendencies of these developments and impact within the operators- scope are also presented.
Abstract: In the modern construction practices, industrial wastes
or by-products are largely used as raw materials in cement and
concrete. These impart many benefits to the environment and bringabout
an economic impact because the cost of waste disposal is
constantly increasing due to strict environmental regulations. It was
reported in literature that the leakage of oil onto concrete element in
older cement grinding unit resulted in concrete with greater resistance
to freezing and thawing. This effect was thought to be similar to
adding an air-entraining chemical admixture to concrete. This paper
presents an investigation on the load deflection behaviour and crack
patterns of reinforced concrete (RC) beams subjected to four point
loading. Ten 120x260x1900 mm beams were cast with 100%
ordinary Portland cement (OPC) concrete, 20% fly ash (FA) and 20%
rice husk ash (RHA) blended cement concrete. 0.15% dosage of
admixtures (used engine oil, new engine oil, and superplasticizer)
was used throughout the experiment. Results show that OPC and
OPC/RHA RC beams containing used engine oil and superplasticizer
exhibit higher capacity, 18-26% than their corresponding control
mix.
Abstract: The choice of studying economics instead of another subject should be motivated by the fact that economics training equips students with skills and knowledge that other disciplines do not provide. Which are these skills and knowledge, however, is not always very clear. This article clarifies such issue by first exploring the philosophical foundations and the defining features of the discipline, and then by investigating in which ways these are transferred to the students. In other words, we study what is meant by the 'economic way of thinking' that is passed on to the students.
Abstract: This paper presents dynamic voltage collapse prediction on an actual power system using support vector machines.
Dynamic voltage collapse prediction is first determined based on the PTSI calculated from information in dynamic simulation output. Simulations were carried out on a practical 87 bus test system by considering load increase as the contingency. The data collected from the time domain simulation is then used as input to the SVM in which support vector regression is used as a predictor to determine the
dynamic voltage collapse indices of the power system. To reduce training time and improve accuracy of the SVM, the Kernel function type and Kernel parameter are considered. To verify the
effectiveness of the proposed SVM method, its performance is compared with the multi layer perceptron neural network (MLPNN). Studies show that the SVM gives faster and more accurate results for dynamic voltage collapse prediction compared with the MLPNN.
Abstract: Proteomics is one of the largest areas of research for
bioinformatics and medical science. An ambitious goal of proteomics
is to elucidate the structure, interactions and functions of all proteins
within cells and organisms. Predicting Protein-Protein Interaction
(PPI) is one of the crucial and decisive problems in current research.
Genomic data offer a great opportunity and at the same time a lot of
challenges for the identification of these interactions. Many methods
have already been proposed in this regard. In case of in-silico
identification, most of the methods require both positive and negative
examples of protein interaction and the perfection of these examples
are very much crucial for the final prediction accuracy. Positive
examples are relatively easy to obtain from well known databases. But
the generation of negative examples is not a trivial task. Current PPI
identification methods generate negative examples based on some
assumptions, which are likely to affect their prediction accuracy.
Hence, if more reliable negative examples are used, the PPI prediction
methods may achieve even more accuracy. Focusing on this issue, a
graph based negative example generation method is proposed, which
is simple and more accurate than the existing approaches. An
interaction graph of the protein sequences is created. The basic
assumption is that the longer the shortest path between two
protein-sequences in the interaction graph, the less is the possibility of
their interaction. A well established PPI detection algorithm is
employed with our negative examples and in most cases it increases
the accuracy more than 10% in comparison with the negative pair
selection method in that paper.
Abstract: Emotion recognition is an important research field that finds lots of applications nowadays. This work emphasizes on recognizing different emotions from speech signal. The extracted features are related to statistics of pitch, formants, and energy contours, as well as spectral, perceptual and temporal features, jitter, and shimmer. The Artificial Neural Networks (ANN) was chosen as the classifier. Working on finding a robust and fast ANN classifier suitable for different real life application is our concern. Several experiments were carried out on different ANN to investigate the different factors that impact the classification success rate. Using a database containing 7 different emotions, it will be shown that with a proper and careful adjustment of features format, training data sorting, number of features selected and even the ANN type and architecture used, a success rate of 85% or even more can be achieved without increasing the system complicity and the computation time
Abstract: This work deals with aspects of support vector machine learning for large-scale data mining tasks. Based on a decomposition algorithm for support vector machine training that can be run in serial as well as shared memory parallel mode we introduce a transformation of the training data that allows for the usage of an expensive generalized kernel without additional costs. We present experiments for the Gaussian kernel, but usage of other kernel functions is possible, too. In order to further speed up the decomposition algorithm we analyze the critical problem of working set selection for large training data sets. In addition, we analyze the influence of the working set sizes onto the scalability of the parallel decomposition scheme. Our tests and conclusions led to several modifications of the algorithm and the improvement of overall support vector machine learning performance. Our method allows for using extensive parameter search methods to optimize classification accuracy.