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.
Abstract: The purpose of the study was to determine the amount of Palm Kernel Oil (PKO) extracted from a packed bed of palm kernels in a supercritical fluid extractor using supercritical carbon dioxide (SC-CO2) as an environmental friendly solvent. Further, the study sought to ascertain the values of the overall mass transfer coefficient (K) of PKO evaluation through a mass transfer model, at constant temperature of 50 °C, 60 °C, and 70 °C and pressures range from 27.6 MPa, 34.5 MPa, 41.4 MPa and 48.3 MPa respectively. Finally, the study also seeks to demonstrate the application of the overall mass transfer coefficient values in relation to temperature and pressure. The overall mass transfer coefficient was found to be dependent pressure at each constant temperature of 50 °C, 60 °C and 70 °C. The overall mass transfer coefficient for PKO in a packed bed of palm kernels was found to be in the range of 1.21X 10-4 m min-1 to 1.72 X 10-4 m min-1 for a constant temperature of 50 °C and in the range of 2.02 X 10-4 m min-1 to 2.43 X 10-4 m min-1 for a constant temperature of 60 °C. Similar increasing trend of the overall mass transfer coefficient from 1.77 X 10-4 m min-1 to 3.64 X 10-4 m min-1 was also observed at constant temperature of 70 °C within the same pressure range from 27.6 MPa to 48.3 MPa.
Abstract: This manuscript presents a method for the numerical solution of the Cauchy type singular integral equations of the first kind, over a finite segment which is bounded at the end points of the finite segment. The Chebyshev polynomials of the second kind with the corresponding weight function have been used to approximate the density function. The force function is approximated by using the Chebyshev polynomials of the first kind. It is shown that the numerical solution of characteristic singular integral equation is identical with the exact solution, when the force function is a cubic function. Moreover, it also shown that this numerical method gives exact solution for other singular integral equations with degenerate kernels.
Abstract: Starting from a biologically inspired framework, Gabor filters were built up from retinal filters via LMSE algorithms. Asubset of retinal filter kernels was chosen to form a particular Gabor filter by using a weighted sum. One-dimensional optimization approaches were shown to be inappropriate for the problem. All model parameters were fixed with biological or image processing constraints. Detailed analysis of the optimization procedure led to the introduction of a minimization constraint. Finally, quantization of weighting factors was investigated. This resulted in an optimized cascaded structure of a Gabor filter bank implementation with lower computational cost.
Abstract: This paper introduces a novel approach to estimate the
clique potentials of Gibbs Markov random field (GMRF) models
using the Support Vector Machines (SVM) algorithm and the Mean
Field (MF) theory. The proposed approach is based on modeling the
potential function associated with each clique shape of the GMRF
model as a Gaussian-shaped kernel. In turn, the energy function of
the GMRF will be in the form of a weighted sum of Gaussian
kernels. This formulation of the GMRF model urges the use of the
SVM with the Mean Field theory applied for its learning for
estimating the energy function. The approach has been tested on
synthetic texture images and is shown to provide satisfactory results
in retrieving the synthesizing parameters.
Abstract: This paper presents a highly efficient algorithm for detecting and tracking humans and objects in video surveillance sequences. Mean shift clustering is applied on backgrounddifferenced image sequences. For efficiency, all calculations are performed on integral images. Novel corresponding exponential integral kernels are introduced to allow the application of nonuniform kernels for clustering, which dramatically increases robustness without giving up the efficiency of the integral data structures. Experimental results demonstrating the power of this approach are presented.
Abstract: Support vector regression (SVR) has been regarded
as a state-of-the-art method for approximation and regression. The
importance of kernel function, which is so-called admissible support
vector kernel (SV kernel) in SVR, has motivated many studies
on its composition. The Gaussian kernel (RBF) is regarded as a
“best" choice of SV kernel used by non-expert in SVR, whereas
there is no evidence, except for its superior performance on some
practical applications, to prove the statement. Its well-known that
reproducing kernel (R.K) is also a SV kernel which possesses many
important properties, e.g. positive definiteness, reproducing property
and composing complex R.K by simpler ones. However, there are a
limited number of R.Ks with explicit forms and consequently few
quantitative comparison studies in practice. In this paper, two R.Ks,
i.e. SV kernels, composed by the sum and product of a translation
invariant kernel in a Sobolev space are proposed. An exploratory
study on the performance of SVR based general R.K is presented
through a systematic comparison to that of RBF using multiple
criteria and synthetic problems. The results show that the R.K is
an equivalent or even better SV kernel than RBF for the problems
with more input variables (more than 5, especially more than 10) and
higher nonlinearity.
Abstract: This paper presents a generalization kernel for gravitational
potential determination by harmonic splines. It was shown
in [10] that the gravitational potential can be approximated using a
kernel represented as a Newton integral over the real Earth body. On
the other side, the theory of geopotential approximation by harmonic
splines uses spherically oriented kernels. The purpose of this paper
is to show that in the spherical case both kernels have the same type
of representation, which leads us to conclusion that it is possible
to consider the kernel represented as a Newton integral over the real
Earth body as a kind of generalization of spherically harmonic kernels
to real geometries.
Abstract: In this paper we present a noise reduction filter for video processing. It is based on the recently proposed two dimensional steering kernel, extended to three dimensions and further augmented to suit the spatial-temporal domain of video processing. Two alternative filters are proposed - the time symmetric kernel and the time asymmetric kernel. The first reduces the noise on single sequences, but to handle the problems at scene shift the asymmetric kernel is introduced. The performance of both are tested on simulated data and on a real video sequence together with the existing steering kernel. The proposed kernels improves the Rooted Mean Squared Error (RMSE) compared to the original steering kernel method on video material.
Abstract: In this paper we introduce a novel kernel classifier
based on a iterative shrinkage algorithm developed for compressive
sensing. We have adopted Bregman iteration with soft and hard
shrinkage functions and generalized hinge loss for solving l1 norm
minimization problem for classification. Our experimental results
with face recognition and digit classification using SVM as the
benchmark have shown that our method has a close error rate
compared to SVM but do not perform better than SVM. We have
found that the soft shrinkage method give more accuracy and in some
situations more sparseness than hard shrinkage methods.
Abstract: Elastic boundary eigensolution problems are converted
into boundary integral equations by potential theory. The kernels of
the boundary integral equations have both the logarithmic and Hilbert
singularity simultaneously. We present the mechanical quadrature
methods for solving eigensolutions of the boundary integral equations
by dealing with two kinds of singularities at the same time. The methods
possess high accuracy O(h3) and low computing complexity. The
convergence and stability are proved based on Anselone-s collective
compact theory. Bases on the asymptotic error expansion with odd
powers, we can greatly improve the accuracy of the approximation,
and also derive a posteriori error estimate which can be used for
constructing self-adaptive algorithms. The efficiency of the algorithms
are illustrated by numerical examples.
Abstract: This paper presents an architecture of current filesystem
implementations as well as our new filesystem SpadFS and operating
system Spad with rewritten VFS layer targeted at high performance
I/O applications. The paper presents microbenchmarks and real-world
benchmarks of different filesystems on the same kernel as well as
benchmarks of the same filesystem on different kernels – enabling
the reader to make conclusion how much is the performance of
various tasks affected by operating system and how much by physical
layout of data on disk. The paper describes our novel features–most
notably continuous allocation of directories and cross-file readahead
– and shows their impact on performance.
Abstract: Steel surface defect detection is essentially one of
pattern recognition problems. Support Vector Machines (SVMs) are
known as one of the most proper classifiers in this application. In this
paper, we introduce a more accurate classification method by using
SVMs as our final classifier of the inspection system. In this scheme,
multiclass classification task is performed based on the "one-againstone"
method and different kernels are utilized for each pair of the
classes in multiclass classification of the different defects.
In the proposed system, a decision tree is employed in the first
stage for two-class classification of the steel surfaces to "defect" and
"non-defect", in order to decrease the time complexity. Based on
the experimental results, generated from over one thousand images,
the proposed multiclass classification scheme is more accurate than
the conventional methods and the overall system yields a sufficient
performance which can meet the requirements in steel manufacturing.
Abstract: Genetic Folding (GF) a new class of EA named as is
introduced for the first time. It is based on chromosomes composed
of floating genes structurally organized in a parent form and
separated by dots. Although, the genotype/phenotype system of GF
generates a kernel expression, which is the objective function of
superior classifier. In this work the question of the satisfying
mapping-s rules in evolving populations is addressed by analyzing
populations undergoing either Mercer-s or none Mercer-s rule. The
results presented here show that populations undergoing Mercer-s
rules improve practically models selection of Support Vector
Machine (SVM). The experiment is trained multi-classification
problem and tested on nonlinear Ionosphere dataset. The target of this
paper is to answer the question of evolving Mercer-s rule in SVM
addressed using either genetic folding satisfied kernel-s rules or not
applied to complicated domains and problems.
Abstract: Text categorization is the problem of classifying text
documents into a set of predefined classes. In this paper, we
investigated three approaches to build a meta-classifier in order to
increase the classification accuracy. The basic idea is to learn a metaclassifier
to optimally select the best component classifier for each
data point. The experimental results show that combining classifiers
can significantly improve the accuracy of classification and that our
meta-classification strategy gives better results than each individual
classifier. For 7083 Reuters text documents we obtained a
classification accuracies up to 92.04%.
Abstract: Text categorization is the problem of classifying text
documents into a set of predefined classes. After a preprocessing
step, the documents are typically represented as large sparse vectors.
When training classifiers on large collections of documents, both the
time and memory restrictions can be quite prohibitive. This justifies
the application of feature selection methods to reduce the
dimensionality of the document-representation vector. In this paper,
three feature selection methods are evaluated: Random Selection,
Information Gain (IG) and Support Vector Machine feature selection
(called SVM_FS). We show that the best results were obtained with
SVM_FS method for a relatively small dimension of the feature
vector. Also we present a novel method to better correlate SVM
kernel-s parameters (Polynomial or Gaussian kernel).
Abstract: In the paper we submit the non-local modification of
kinetic Smoluchowski equation for binary aggregation applying to
dispersed media having memory. Our supposition consists in that that
intensity of evolution of clusters is supposed to be a function of the
product of concentrations of the lowest orders clusters at different
moments. The new form of kinetic equation for aggregation is
derived on the base of the transfer kernels approach. This approach
allows considering the influence of relaxation times hierarchy on
kinetics of aggregation process in media with memory.
Abstract: The biological activity of A. pullulans isolates against
species of the genus Fusarium, bacteria of the genus Azotobacter and
pseudomonads colonizing wheat kernels was evaluated. A field
experiment was carried out in 2009-2011, in north-eastern Poland.
Winter wheat (cv. Bogatka) plants were sprayed with a cell
suspension of A. pullulans at a density of 106 - 108 per cm3 water at
the stem elongation stage and the heading stage. Untreated plants
served as control. The abundance of epiphytic yeasts, bacteria of the
genus Azotobacter, pseudomonads and Fusarium pathogens on wheat
grain was estimated at harvest and after six months’ storage. The
average size of yeast communities was significantly greater on wheat
kernels treated with a cell suspension of A. pullulans, compared with
control samples. In 2010-2011, biological control reduced the
abundance of some species of the genus Fusarium.
Abstract: Understanding proteins functions is a major goal in
the post-genomic era. Proteins usually work in context of other
proteins and rarely function alone. Therefore, it is highly relevant to
study the interaction partners of a protein in order to understand its
function. Machine learning techniques have been widely applied to
predict protein-protein interactions. Kernel functions play an
important role for a successful machine learning technique. Choosing
the appropriate kernel function can lead to a better accuracy in a
binary classifier such as the support vector machines. In this paper,
we describe a Bayesian kernel for the support vector machine to
predict protein-protein interactions. The use of Bayesian kernel can
improve the classifier performance by incorporating the probability
characteristic of the available experimental protein-protein
interactions data that were compiled from different sources. In
addition, the probabilistic output from the Bayesian kernel can assist
biologists to conduct more research on the highly predicted
interactions. The results show that the accuracy of the classifier has
been improved using the Bayesian kernel compared to the standard
SVM kernels. These results imply that protein-protein interaction can
be predicted using Bayesian kernel with better accuracy compared to
the standard SVM kernels.
Abstract: Speedups from mapping four real-life DSP
applications on an embedded system-on-chip that couples coarsegrained
reconfigurable logic with an instruction-set processor are
presented. The reconfigurable logic is realized by a 2-Dimensional
Array of Processing Elements. A design flow for improving
application-s performance is proposed. Critical software parts, called
kernels, are accelerated on the Coarse-Grained Reconfigurable
Array. The kernels are detected by profiling the source code. For
mapping the detected kernels on the reconfigurable logic a prioritybased
mapping algorithm has been developed. Two 4x4 array
architectures, which differ in their interconnection structure among
the Processing Elements, are considered. The experiments for eight
different instances of a generic system show that important overall
application speedups have been reported for the four applications.
The performance improvements range from 1.86 to 3.67, with an
average value of 2.53, compared with an all-software execution.
These speedups are quite close to the maximum theoretical speedups
imposed by Amdahl-s law.