Downloads & Free Reading Options - Results

Deep neural networks have been exhibiting splendid accuracies in many of visual pattern classification problems. Many of the state-of-the-art methods employ a technique known as data augmentation at the training stage. This paper addresses an issue of decision rule for classifiers trained with augmented data. Our method is named as APAC: the Augmented PAttern Classification, which is a way of classification using the optimal decision rule for augmented data learning. Discussion of methods of data augmentation is not our primary focus. We show clear evidences that APAC gives far better generalization performance than the traditional way of class prediction in several experiments. Our convolutional neural network model with APAC achieved a state-of-the-art accuracy on the MNIST dataset among non-ensemble classifiers. Even our multilayer perceptron model beats some of the convolutional models with recently invented stochastic regularization techniques on the CIFAR-10 dataset.

A massively parallel architecture for pattern classification is described. The architecture is based on the field of density estimation. It makes use of a variant of the adaptive kernel estimator to approximate the distributions of the classes as a sum of Gaussian distributions. These Gaussians are learned using a moved mean, moving covariance learning scheme. A temporal ordering scheme is implemented using decay at the input level, allowing the network to learn to recognize sequences. The learning scheme requires a single pass through the data, giving the architecture the capability of real time learning. The first part of the paper develops the adaptive kernel estimator. The parallel architecture is then described, and issues relevant to implementation are discussed. Finally, applications to robotic sensor fusion, intended word recognition, and vision are described. Keywords: Gaussian distributions, Density estimation, Pattern recognition.

This article is from Bioinformatics , volume 28 . Abstract Motivation: Knowledge of the subcellular location of a protein is crucial for understanding its functions. The subcellular pattern of a protein is typically represented as the set of cellular components in which it is located, and an important task is to determine this set from microscope images. In this article, we address this classification problem using confocal immunofluorescence images from the Human Protein Atlas (HPA) project. The HPA contains images of cells stained for many proteins; each is also stained for three reference components, but there are many other components that are invisible. Given one such cell, the task is to classify the pattern type of the stained protein. We first randomly select local image regions within the cells, and then extract various carefully designed features from these regions. This region-based approach enables us to explicitly study the relationship between proteins and different cell components, as well as the interactions between these components. To achieve these two goals, we propose two discriminative models that extend logistic regression with structured latent variables. The first model allows the same protein pattern class to be expressed differently according to the underlying components in different regions. The second model further captures the spatial dependencies between the components within the same cell so that we can better infer these components. To learn these models, we propose a fast approximate algorithm for inference, and then use gradient-based methods to maximize the data likelihood.Results: In the experiments, we show that the proposed models help improve the classification accuracies on synthetic data and real cellular images. The best overall accuracy we report in this article for classifying 942 proteins into 13 classes of patterns is about 84.6%, which to our knowledge is the best so far. In addition, the dependencies learned are consistent with prior knowledge of cell organization.Availability:http://murphylab.web.cmu.edu/software/.Contact:Jeff.Sch [email protected], [email protected]

A methodology based on the concept of variable string length GA (VGA) is developed for determining automatically the number of hyperplanes for modeling the class boundaries in GA classifier. The genetic operators and fitness function are newly defined to take care of the variability in chromosome length. It is proved that the said method is able to arrive at the optimal number of misclassifications after sufficiently large number of iterations, and will need minimal number of hyperplanes for this purpose. Experimental results on different artificial and real life data sets demonstrate that the classifier, using the concept of variable length chromosome, can automatically evolve an appropriate value of H, and also provide performance better than those of the fixed length version. Its comparison with another approach using VGA is provided.

A separation algorithm applicable to the pattern classification and cluster analysis of n-dimensional (n > 2) data is presented. The algorithm reduces the dimensionality of the problem by projecting each point into a plane. This plane is presented to the user on a computer graphics console screen. The operator picks a point on the screen with a lightpen and chooses a \"direction of movement\" to achieve or increase separation, thereby causing an iteration of the algorithm. Each iteration is in fact a reorientation of the plane into which the data points are projected. Iterations continue until satisfactory separation is achieved. The algorithm is not restricted by the dimensionality of the data, nor are any distributional assumptions required. Results from six case studies indicate that the algorithm is a useful tool for the analysis of multidimensional data.

This research introduces the appropriate input pattern of Fuzzy Logic design for fault type classification of Single Line to Ground Fault at distribution network. The proposed design is solely using Fuzzy Logic as the research technique with input data from PSCAD simulation. PSCAD software simulate the circuit configuration for fault disturbance at the distribution network. The research technique was applied with multiples input values of voltage and current that extracted from the PSCAD simulation. This research testifies the output result by using different fault resistance values; 0.01Ω, 10Ω, 30Ω, 50Ω and 70Ω. Voltage sag and current swell of phase a, b and c that were obtained from the PSCAD simulation have been used as the input variables for Fuzzy Logic design. The acquired results that represented in average accuracy shown that voltage sag and current swell can draw a satisfying accuracy in classifying the fault type. 

This research introduces the appropriate input pattern of Fuzzy Logic design for fault type classification of Single Line to Ground Fault at distribution network. The proposed design is solely using Fuzzy Logic as the research technique with input data from PSCAD simulation. PSCAD software simulate the circuit configuration for fault disturbance at the distribution network. The research technique was applied with multiples input values of voltage and current that extracted from the PSCAD simulation. This research testifies the output result by using different fault resistance values; 0.01Ω, 10Ω, 30Ω, 50Ω and 70Ω. Voltage sag and current swell of phase a, b and c that were obtained from the PSCAD simulation have been used as the input variables for Fuzzy Logic design. The acquired results that represented in average accuracy shown that voltage sag and current swell can draw a satisfying accuracy in classifying the fault type. 

A pattern classification scheme which is grounded in classical probability theory may be associated with confidence internvals that represent an estimate of the predictive capability of the scheme. As a practical matter, realistic allocations of data acquisition and processing resources may severely constrain acceptable levels of predictability. We examine some of the basic assumptions which underlie the standard statistical techniques. In particular, we show that fuzzy logic effectively produces conservative estimates for the conditional probability of the union of sets since, in that case, it neglects information related to the intersection. We propose that such neglect can be remedied, at a computational cost, without resorting explicitly to the usual procedure of integrating over irregularly shaped volumes. To this end, we introduce a class of probability density distributions which possesses (hyper) rectangular contour. Explicit formulas for the normalization constant and the probability of error are then derived for typical distributions.

Infrared spectroscopy is an important technique for measuring airborne chemicals, for pollution monitoring and to warn of toxic compound releases. Infrared spectroscopy provides both detection and identification of airborne components. Computer-assisted classification tools, including pattern recognition and artificial neural network techniques, have been applied to a collection of infrared spectra of organophosphorus compounds, and these have successfully discriminated commercial pesticide compounds from military nerve agents, precursors, and hydrolysis products. Infrared spectra for previous tests came from a commercial infrared library, with permission, from military laboratories, and from defense contractors. In order to further test such classification tools, additional infrared spectra from the NIST gas-phase infrared library were added to the data set. These additional spectra probed the tendency of the trained classifiers to misidentify unrelated spectra into the trained classes. Infrared spectra used in this effort were gathered from a variety of sources. Different instrument operators collected them at a number of locations, in a variety of spectral data collection designs, and they were delivered in a variety of digital formats. The spectra were treated mathematically to remove artifacts from their collection. Preprocessing techniques used included Fisher weighting and principal component analysis. Classifications were made using the k-nearest neighbor classifier, feed forward neural networks, trained with a variety of techniques, and radial basis function networks. The results from these classification techniques will be reported and compared.

Information systems are often inundated with thousands of attack alerts to distinguish novice hacker probes from genuine threats. Pattern classification can help filter relatively benign attacks from alerts generated by anomaly detectors, limited the numbers of alerts to requiring attention. This research investigates the feasibility of using pattern classification algorithms on network packed header information to classify network attacks. Both liner discrimination and radial basis function algorithms are trained using flood and scan attacks. The classifiers are then tested with unknown floods and scans to determine how well they categorize previously unseen attacks.

Information systems are often inundated with thousands of attack alerts to distinguish novice hacker probes from genuine threats. Pattern classification can help filter relatively benign attacks from alerts generated by anomaly detectors, limited the numbers of alerts to requiring attention. This research investigates the feasibility of using pattern classification algorithms on network packed header information to classify network attacks. Both liner discrimination and radial basis function algorithms are trained using flood and scan attacks. The classifiers are then tested with unknown floods and scans to determine how well they categorize previously unseen attacks.

The study examines the rule space model, a probabilistic model capable of measuring cognitive skill acquisition and of diagnosing erroneous rules of operation in a procedural domain. The model involves two important components: (1) determination of a set of bug distributions (bug density functions representing clusters around the rules); and (2) establishment of decision rules for classifying an observed response pattern into one of the clusters around the rules and computing error probabilities. This study further discusses the theoretical foundation of the model by introducing "bug distribution" and hypothesis testing (Bayes' decision rules for minimum errors). The model does not require a large-scale computation. It is helpful in areas of research and teaching. It can be used with microcomputers for testing hypotheses and validating data with probabilistically-sound information, and it can improve and modify information for the cluster ellipses as more students are added to research projects. (LMO)

The multi-layer perceptron (MLP) is a type of adaptive layered network often used as a pattern classifier. In more recent literature, MLPs are compared with simpler classification techniques using common datasets. We select two of these simple static pattern classification algorithms and briefly review the relevant techniques. After introducing a modest set of evaluation databases, the performance of the standard classifiers and MLPs are assessed. A technique for implementing the two standard classifiers as MLPs is presented and this novel approach is used to automatically design a 'good' set of initial weights for the MLP networks. Encouraging experimental results for these hybrid techniques are shown for illustration. Keywords: Pattern recognition; Speech; Images; Artificial intelligence.

The multi-layer perceptron (MLP) is a type of adaptive layered network often used as a pattern classifier. In more recent literature, MLPs are compared with simpler classification techniques using common datasets. We select two of these simple static pattern classification algorithms and briefly review the relevant techniques. After introducing a modest set of evaluation databases, the performance of the standard classifiers and MLPs are assessed. A technique for implementing the two standard classifiers as MLPs is presented and this novel approach is used to automatically design a 'good' set of initial weights for the MLP networks. Encouraging experimental results for these hybrid techniques are shown for illustration. Keywords: Pattern recognition; Speech; Images; Artificial intelligence.

This article is from Journal of Neurophysiology , volume 110 . Abstract The generation of a behaviorally relevant cue to the speed of objects around us is critical to our ability to navigate safely within our environment. However, our perception of speed is often distorted by prevailing conditions. For instance, as luminance is reduced, our perception of the speed of fast-moving patterns can be increased by as much as 30%. To investigate how the cortical representation of speed may vary under such conditions, we have measured the functional MRI blood oxygen level-dependent (BOLD) response of visual cortex to drifting sine gratings at two very different luminances. The average BOLD response in all areas was band-pass with respect to speed (or equivalently, temporal frequency) and thus contained no unambiguous speed information. However, a multivariate classifier was able to predict grating speed successfully in all cortical areas measured. Similarly, we find that a multivariate classifier can predict stimulus luminance. No differences in either the mean BOLD response or the multivariate classifier response with respect to speed were found as luminance changed. However, examination of the spatial distribution of speed preferences in the primary visual cortex revealed that perifoveal locations preferred slower speeds than peripheral locations at low but not high luminance. We conclude that although an explicit representation of perceived speed has yet to be demonstrated in the human brain, multiple visual regions encode both the temporal structure of moving stimuli and luminance implicitly.

Abstract: In this paper, a new approach for facial gender classification is presented. The human face serves as a knowledge base for useful demographic information such as gender, expression and age. We can easily identify the gender of the person, but it is difficult for the machine to recognize gender from the face image. The face is a complex three dimensional object and it is a challenging task for a machine to recognise gender due to a wide degree of variations in texture and shape of the face. The gender recognition has many … more Keywords: Face Gender, local directional pattern, Princiapl Coomponets Analysis , Local Binary Pattern Authors: Mr. N. K. Bansode, AIT Prof. Dr. P. K. Sinha, AIT Dates of Issue (DOI) Submission Date: Wed, 27. Apr. 2016 Review Date: Wed, 11. May. 2016 Publishing Date: Wed, 1. Jun. 2016 http://www.icgst.com/paper.aspx?pid=P1151618493

Abstract: In this paper, a new approach for facial gender classification is presented. The human face serves as a knowledge base for useful demographic information such as gender, expression and age. We can easily identify the gender of the person, but it is difficult for the machine to recognize gender from the face image. The face is a complex three dimensional object and it is a challenging task for a machine to recognise gender due to a wide degree of variations in texture and shape of the face. The gender recognition has many … more Keywords: Face Gender, local directional pattern, Princiapl Coomponets Analysis , Local Binary Pattern Authors: Mr. N. K. Bansode, AIT Prof. Dr. P. K. Sinha, AIT Dates of Issue (DOI) Submission Date: Wed, 27. Apr. 2016 Review Date: Wed, 11. May. 2016 Publishing Date: Wed, 1. Jun. 2016 http://www.icgst.com/paper.aspx?pid=P1151618493

Patterndassification and clustering algorithms are key components of modern information processing systems used to perform tasks such as speech and image recognition, printedcharacter recognition, medical diagnosis, fault detection, process control, and financial decision making. To simplifY the task of applying these types of algorithms in new application areas, we have developed LNKnet-a software package that provides access toinore than 20 patternclassification, clustering, and featureselection algorithms. Included are the most important algorithms from the fields of neural networks, statistics, machine learning, and artificial intelligence. The algorithms can be trained and tested on separate data or tested with automatic crossvalidation. LNKnet runs under the UNIX operating system and access to the different algorithms is provided through a graphical pointandclick user interface. Graphical outputs include twodimensional (2D) scatter and decisionregion plots and 1-D plots of data histograms, classifier outputs, and error rates during training. Parameters of trained classifiers are stored in files from which the parameters can be translated into source-code subroutines (written in the C programming language) that can then be embedded in a user application program. Lincoln Laboratory and other research laboratories have used LNKnet successfully for many diverse applications.

The author has identified the following significant results. The sand beach was separated from sea water in each of four bands, if the beach had a width of 100 m or more. Density ranges of the sea for CCT counts were determined as 0-3 for band 7, 0-16 for band 6, 0-25 for band 5, and 0-27 for band 4.

Preregistration of Hypothesis in the CoScience Project.

Preregistration of Hypothesis in the CoScience Project.

Preregistration of Hypothesis in the CoScience Project.

Preregistration of Hypothesis in the CoScience Project.

Neuromorphic pattern classifiers were implemented, for the first time, using transistor-free integrated crossbar circuits with bilayer metal-oxide memristors. 106- and 108-crosspoint neuromorphic networks were trained in-situ using a Manhattan-Rule algorithm to separate a set of 33 binary images: into 3 classes using the batch-mode training, and into 4 classes using the stochastic-mode training, respectively. Simulation of much larger, multilayer neural network classifiers based on such technology has shown that their fidelity may be on a par with the state-of-the-art results.

Neuromorphic pattern classifiers were implemented, for the first time, using transistor-free integrated crossbar circuits with bilayer metal-oxide memristors. 106- and 108-crosspoint neuromorphic networks were trained in-situ using a Manhattan-Rule algorithm to separate a set of 33 binary images: into 3 classes using the batch-mode training, and into 4 classes using the stochastic-mode training, respectively. Simulation of much larger, multilayer neural network classifiers based on such technology has shown that their fidelity may be on a par with the state-of-the-art results.

Infrared spectra of organophosphorus compounds, including pesticides and a set of neurotoxins which have been banned from use by international agreement, along with their precursors and hydrolysis products, were obtained from a variety of sources. The data were treated to minimize spectral information related to the spectral origin. A common spectral wavelength range was selected and spectral data within this range were transduced into data vectors. Computer-assisted classification tools were used to classify the spectra as pesticides versus neurotoxins and their precursors and hydrolysis products. The performance of a k-nearest neighbor classifier for this distinction is compared with several artificial neural network classifiers.

We consider the statistical problem of catalogue matching from a machine learning perspective with the goal of producing probabilistic outputs, and using all available information. A framework is provided that unifies two existing approaches to producing probabilistic outputs in the literature, one based on combining distribution estimates and the other based on combining probabilistic classifiers. We apply both of these to the problem of matching the HIPASS radio catalogue with large positional uncertainties to the much denser SuperCOSMOS catalogue with much smaller positional uncertainties. We demonstrate the utility of probabilistic outputs by a controllable completeness and efficiency trade-off and by identifying objects that have high probability of being rare. Finally, possible biasing effects in the output of these classifiers are also highlighted and discussed.

A large class of fractional quantum Hall (FQH) states can be classified according to their pattern of zeros, which describes the way ideal ground state wave functions go to zero as various clusters of electrons are brought together. In this paper we generalize this approach to classify multilayer FQH states. Such a classification leads to the construction of a class of non-Abelian multilayer FQH states that are closely related to $\hat{g}_k$ parafermion conformal field theories, where $\hat{g}_k$ is an affine simple Lie algebra. We discuss the possibility of some of the simplest of these non-Abelian states occuring in experiments on bilayer FQH systems at $\nu = 2/3$, 4/5, 4/7, etc.

http://uf.catalog.fcla.edu/uf.jsp?st=UF000687925&ix=nu&I=0&V=D

Electrocardiogram (ECG) reflects activity of the central of the blood circulatory system, i.e. the heart. An ECG signal can provide us with a great deal of information on the normal and pathological physiology of heart activity. Thus, ECG is an important non-invasive clinical tool for the diagnosis of heart diseases.          According to the medical definition the most important information in the ECG signal is concentrated in the P wave, QRS complex and T wave. These data include positions and/or magnitudes of the QRS interval, PR interval, QT interval, ST interval, PR segment, and ST segment (see Fig. 1). Based on the above data, doctors can correctly diagnose human heart diseases. Therefore, analyzing the ECG signals of cardiac arrhythmia is very important for doctors to make correct clinical diagnoses. In order to perform ECG signals classification of the cardiac arrhythmia, the first important task is to determine an appropriate set of features. The feature selection method which chooses the best features from original features to have the maximum recognition rate, simplify classified computation and comprehend the causal relation of classified question. Signal Processing   is undoubtedly the best real time implementation of a specific problem. Wavelet Transform is a very powerful technique for feature extraction and can be used along with neural network structures to build computationally efficient models for diagnosis of Biosignals (ECG in this case). This work utilizes the above techniques for diagnosis of an ECG signal by determining its nature as well as exploring the possibility for real-time implementation of the above model. Daubechies wavelet transform and multi-layered perceptron are the computational techniques used for the realization of the above model. The ECG signals were obtained from the MIT-BIH arrhythmia database and are used for the identification of four different types of arrhythmias. The identification was implemented real-time in SIMULINK, to simulate the detection model under test condition and verify its workability.

Electrocardiogram (ECG) reflects activity of the central of the blood circulatory system, i.e. the heart. An ECG signal can provide us with a great deal of information on the normal and pathological physiology of heart activity. Thus, ECG is an important non-invasive clinical tool for the diagnosis of heart diseases.          According to the medical definition the most important information in the ECG signal is concentrated in the P wave, QRS complex and T wave. These data include positions and/or magnitudes of the QRS interval, PR interval, QT interval, ST interval, PR segment, and ST segment (see Fig. 1). Based on the above data, doctors can correctly diagnose human heart diseases. Therefore, analyzing the ECG signals of cardiac arrhythmia is very important for doctors to make correct clinical diagnoses. In order to perform ECG signals classification of the cardiac arrhythmia, the first important task is to determine an appropriate set of features. The feature selection method which chooses the best features from original features to have the maximum recognition rate, simplify classified computation and comprehend the causal relation of classified question. Signal Processing   is undoubtedly the best real time implementation of a specific problem. Wavelet Transform is a very powerful technique for feature extraction and can be used along with neural network structures to build computationally efficient models for diagnosis of Biosignals (ECG in this case). This work utilizes the above techniques for diagnosis of an ECG signal by determining its nature as well as exploring the possibility for real-time implementation of the above model. Daubechies wavelet transform and multi-layered perceptron are the computational techniques used for the realization of the above model. The ECG signals were obtained from the MIT-BIH arrhythmia database and are used for the identification of four different types of arrhythmias. The identification was implemented real-time in SIMULINK, to simulate the detection model under test condition and verify its workability.

This project was completed to determine if fuzzy set classification could be used to accurately evaluate data collected during a psychophysiological detection of deception examination. This methodology provides an alternative to the proprietary statistical technique now commonly used. Data collected using both the Modified General Question Technique (MGQT) and the Relevant Only formats were evaluated. An extensive and, arguably, complete set of polygraph data features was identified. These polygraph data features were not individual dependent, examiner dependent, or in any way dependent on apriori or posteriori knowledge (statistics) of the data. A fuzzy K-Nearest Neighbor classifier and an adaptive fuzzy Least Mean Squares classifier were developed. A fuzzy C-Means clustering algorithm which enabled visualization of the data features was also developed. The fuzzy algorithms were 'forced' to make a choice of truth versus deception; they could, however, be used to return a number that would, in near real-time, give the examiner an idea of the confidence level of the algorithm. The data were parsed such that 25% of the data were tested using an algorithm developed from the remaining 75% of the data. It is shown that only four features are needed to achieve 100% correct classification of the Relevant Only data and 97% correct classification of the MGQT data. It is suggested that any future research development, or testing or computer classification techniques, including statistical and neural techniques include the results of this work.

Point patterns are sets or multi-sets of unordered elements that can be found in numerous data sources. However, in data analysis tasks such as classification and novelty detection, appropriate statistical models for point pattern data have not received much attention. This paper proposes the modelling of point pattern data via random finite sets (RFS). In particular, we propose appropriate likelihood functions, and a maximum likelihood estimator for learning a tractable family of RFS models. In novelty detection, we propose novel ranking functions based on RFS models, which substantially improve performance.

The missing data in incomplete pattern can have different estimations, and the classification result of pattern with different estimations may be quite distinct. Such uncertainty (ambiguity) of classification is mainly caused by the loss of information in missing data. A new prototype-based credal classification (PCC) method is proposed to classify incomplete patterns using belief functions. The class prototypes obtained by the training data are respectively used to estimate the missing values. Typically, in a c-class problem, one has to deal with c prototypes which yields c estimations. The different edited patterns based on each possible estimation are then classified by a standard classifier and one can get c classification results for an incomplete pattern. Because all these classification results are potentially admissible, they are fused altogether to obtain the credal classification of the incomplete pattern. A new credal combination method is introduced for solving the classification problem, and it is able to characterize the inherent uncertainty due to the possible conflicting results delivered by the different estimations of missing data. The incomplete patterns that are hard to correctly classify will be reasonably committed to some proper meta-classes by PCC method in order to reduce the misclassification rate. The use and potential of PCC method is illustrated through several experiments with artificial and real data sets.

The use of patterns in predictive models is a topic that has received a lot of attention in recent years. Pattern mining can help to obtain models for structured domains, such as graphs and sequences, and has been proposed as a means to obtain more accurate and more interpretable models. Despite the large amount of publications devoted to this topic, we believe however that an overview of what has been accomplished in this area is missing. This paper presents our perspective on this evolving area. We identify the principles of pattern mining that are important when mining patterns for models and provide an overview of pattern-based classification methods. We categorize these methods along the following dimensions: (1) whether they post-process a pre-computed set of patterns or iteratively execute pattern mining algorithms; (2) whether they select patterns model-independently or whether the pattern selection is guided by a model. We summarize the results that have been obtained for each of these methods.

The use of patterns in predictive models is a topic that has received a lot of attention in recent years. Pattern mining can help to obtain models for structured domains, such as graphs and sequences, and has been proposed as a means to obtain more accurate and more interpretable models. Despite the large amount of publications devoted to this topic, we believe however that an overview of what has been accomplished in this area is missing. This paper presents our perspective on this evolving area. We identify the principles of pattern mining that are important when mining patterns for models and provide an overview of pattern-based classification methods. We categorize these methods along the following dimensions: (1) whether they post-process a pre-computed set of patterns or iteratively execute pattern mining algorithms; (2) whether they select patterns model-independently or whether the pattern selection is guided by a model. We summarize the results that have been obtained for each of these methods.

This report describes the application of 10 pattern classification techniques to selected samples of high resolution, synthetic aperture radar imagery taken over the Huntsville, Alabama area. Sections of the radar imagery were digitized and stored on a digital disk unit. A Lexidata system 3400 image processor and a Hewlett Packard 1000 computer were used to display the images on a cathode ray tube and to take 100 samples for each of four terrain classes from the imagery. The 400 image samples were then used as training sets to derive the 10 classifiers. Once the classifiers were derived, the training set data were then used as input to the classifiers to see how well each would do in classifying the original training sets.

Thesis (MS)?Naval Postgraduate School, 1964

Synthetic aperture radar systems have made possible the generation of radar images of ships with high enough resolution to allow numerous targets or scatterers to be visible. With the availability of numerous scatterers in one radar image, it is theoretically possible to identify the class of a ship. The SAR image of a ship is a function of location of scatterers, SAR system frequency, radar-to-ship viewing angle, amount and type of sea-induced ship motion, and length of aperture. Because of the dependence on these variables, the number of images representing any one ship is large. It is the job of radar operator to study and understand the many radar images that can be encountered, and attempt to make the correct classification. Fast classification response times are required, since these images would normally be acquired in real time. A syntactic pattern recognition algorithm (called the Coarse Feature Classifier (CFC)) has been developed to aid the radar operator to perform the task of classifying SAR images of ships. By having the algorithm perform some of the tasks that the operator normally performs, one obtains the potential benefits of improved accuracy and speed of classification, and reduced operator fatigue. The algorithm extracts numerous features from the input SAR image which are then compared to a library of similar features in order to select the ship(s) from the library which best resembles the input ship image. Details of the operation of the CFC are discussed.

This is the final report of a two year study on the use of fuzzy pattern recognition of polygraph data for the identification of truth versus deception. The goals of this study as stated in the original proposal where to: (1) develop a data parsing algorithm which will process polygraph data obtained from the NSA into three domains: time-domain, frequency domain, and correlation domain; (2) design a frizzy classifier algorithm to accept the featurized data and modify its membership functions based on the error between its classification of the polygraph data and the classification in the NSA files; (3) study relationship between number of membership functions an the success of the data classification and; (4) investigate the feasibility of the classification being performed in a near-real-time scenario. The data to be used was MGQT polygraph data. However, the proposal for the second year of the study introduced the goal of comparing the performance of the developed frizzy classification system with 'zone comparison' polygraph data. Ultimately this was changed to be the simulated 'relevant only' data obtained from DODPI. There were two secondary objectives of this project. First, are the features identified as optimal in determining the veracity of a subject optimal for all subjects. Second, are there features not presently being used in polygraph analysis the may be optimal. This report and its attached appendices will show that all objectives of the original proposal where met. A frizzy parser and classifier system were developed that could run in near real-time, achieve performances as good or better than the presently available automatic polygraph systems, and identify new features that previously where not used in polygraph classification. Results of 97% correct for the MGQT data and 100% correct for the 'relevant only' data were achieved.

This report describes progress in the development of the area classification portion of a computer vision system for cloud pattern analysis. The ultimate goal of the vision system is to extract meteorologically significant cloud regions from a time sequence of dual-channel geosynchronous satellite images. The question explored by this paper is to what extend single-stage and multistage statistical pattern recognition techniques may be employed in the classification of clouds from a single dual-channel image. (Author)

This report describes progress in the development of the area classification portion of a computer vision system for cloud pattern analysis. The ultimate goal of the vision system is to extract meteorologically significant cloud regions from a time sequence of dual-channel geosynchronous satellite images. The question explored by this paper is to what extend single-stage and multistage statistical pattern recognition techniques may be employed in the classification of clouds from a single dual-channel image. (Author)

In this study we describe the development of a ride assistance application which can be implemented on the widespread smart phones and tablet. The ride assistance application has a signal processing and pattern classification module which yield almost 100% recognition accuracy for real-time signal pattern classification. We introduce a novel framework to build a training dictionary with an overwhelming discriminating capacity which eliminates the need of human intervention spotting the pattern on the training samples. We verify the recognition accuracy of the proposed methodologies by providing the results of another study in which the hand posture and gestures are tracked and recognized for steering a robotic wheelchair.

In this study we describe the development of a ride assistance application which can be implemented on the widespread smart phones and tablet. The ride assistance application has a signal processing and pattern classification module which yield almost 100% recognition accuracy for real-time signal pattern classification. We introduce a novel framework to build a training dictionary with an overwhelming discriminating capacity which eliminates the need of human intervention spotting the pattern on the training samples. We verify the recognition accuracy of the proposed methodologies by providing the results of another study in which the hand posture and gestures are tracked and recognized for steering a robotic wheelchair.

A method is described for finding decision boundaries, approximated by piecewise linear segments, for classifying patterns in RN N >/= 2, using simulated annealing. It involves generation and placement of a set of hyperplanes (represented by strings) in the feature space that yields minimum misclassification. Theoretical analysis shows that as the size of the training data set approaches infinity, the boundary provided by the simulated annealing based classifier will approach the Bayes boundary. The effectiveness of the classification methodology, along with the generalization ability of the decision boundary, is demonstrated for both artificial data and real life data sets having non-linear/overlapping class boundaries. Results are compared extensively with those of the Bayes classifier, k-NN rule and multilayer perceptron, and Genetic Algorithms, another popular evolutionary technique. Empirical verification of the theoretical claim is also provided.

One method for classifying natural fracture systems is by fracture genesis. This approach involves the physics of the formation process, and it has been used most frequently in attempts to predict subsurface fractures and petroleum reservoir productivity. This classification system can also be applied to larger fracture systems on any planetary surface. One problem in applying this classification system to planetary surfaces is that it was developed for ralatively small-scale fractures that would influence porosity, particularly as observed in a core sample. Planetary studies also require consideration of large-scale fractures. Nevertheless, this system offers some valuable perspectives on fracture systems of any size.

In two-class pattern recognition, it is a standard technique to have an algorithm for finding hyperplanes which separates the two classes in a linearly separable training set. The traditional methods find a hyperplane which separates all points in one class from all point in the other, but such a hyperplane is not necessarily centered in the empty space between the two classes. Since a central hyperplane does not favor one class or the other, it should have a lower error rate in classifying new points and is therefore better than a noncentral hyperplane. Six algorithms for finding central hyperplanes are tested on three data sets. Although frequently used in practice, the modified relaxation algorithm is very poor. Three algorithms, which are defined in the paper, are found to be quite good. (Author)

This paper introduces a probabilistic model that is capable of diagnosing and classifying cognitive errors in a general problem-solving domain. The model is different from the usual deterministic strategies common in the area of artificial intelligence because the item response theory is utilized for handling the variability of response errors. As for illustrating the model, the dataset obtained form a 38-item fraction addition test is used, and the students' responses are classified into 34 groups of misconceptions. These groups are predetermined by the result of an error analysis previously done, and validated with the error diagnostic program written by a typical formal logic approach. Keywords: cognitive errors, item response theory, bugs, fractions, pattern classification, caution index.

Multiphoton femtosecond coherent control of is used for implementing innovative photo-induced analog coherent computation that generally might be a basis for future "smart hardware". The specific implemented computational task the classification of an unknown sequence into one of the three groups: (i) a constant sequence that is composed of identical numbers, (ii) a sequence that is antisymmetric around a given point, or (iii) neither. The input sequence is encoded into the spectral phases of a broadband femtosecond pulse and the computational task is being carried out by the multiphoton nonlinear response of the irradiated physical system. Here, it is the simultaneous coherent two- and three-photon absorption in atomic sodium (Na). The corresponding computational resources are the manifold of initial-to-final multiphoton excitation pathways photo-induced by the broad spectrum of the irradiating femtosecond pulse. The answer is obtained by measuring only two observables, which are two state-populations excited via the two- and three-photon absorption processes. Hence, the computational task is accomplished in a constant number of operations irrespective of the sequence length. As such, the presented scheme, where the femtosecond pulse serves as a query and the irradiated physical system serves as an oracle, provides a sufficient gain in the computational complexity.

The various studies conducted for classification of handwritten signatures of people have shown that the task is difficult because there is intra personal differences among the signatures of the same person. The signatures of the same person vary with time, age of the person and also because of the emotional state of a person.