WORD SENSE DISAMBIGUATION BASED ON IMPROVED BAYESIAN CLASSIFIERS

Word Sense Disambiguation (WSD) is to decide the sense of an ambiguous word on particular context. Most of current studies on WSD only use several ambiguous words as test samples, thus leads to some limitation in practical application. In this paper, we perform WSD study based on large scale real-world corpus using two unsupervised learning algorithms based on ±n-improved Bayesian model and Dependency Grammar (DG)-improved Bayesian model. ±n-improved classifiers reduce the window size of context of ambiguous words with close-distance feature extraction method, and decrease the jamming of useless features, thus obviously improve the accuracy, reaching 83.18% (in open test). DG-improved classifier can more effectively conquer the noise effect existing in Naive-Bayesian classifier. Experimental results show that this approach does better on Chinese WSD, and the open test achieved an accuracy of 86.27%.

A new constrained maximum margin approach to discriminative learning of Bayesian classifiers

We propose a novel discriminative learning approach for Bayesian pattern classification, called ＇constrained maximum margin （CMM）＇. We define the margin between two classes as the difference between the minimum decision value for positive samples and the maximum decision value for negative samples. The learning problem is to maximize the margin under the con- straint that each training pattern is classified correctly. This nonlinear programming problem is solved using the sequential un- constrained minimization technique. We applied the proposed CMM approach to learn Bayesian classifiers based on Gaussian mixture models, and conducted the experiments on 10 UCI datasets. The performance of our approach was compared with those of the expectation-maximization algorithm, the support vector machine, and other state-of-the-art approaches. The experimental results demonstrated the effectiveness of our approach.

Adaptive target and jamming recognition for the pulse doppler radar fuze based on a time-frequency joint feature and an online-updated naive bayesian classifier with minimal risk

This paper considers the problem of target and jamming recognition for the pulse Doppler radar fuze(PDRF).To solve the problem,the matched filter outputs of the PDRF under the action of target and jamming are analyzed.Then,the frequency entropy and peak-to-peak ratio are extracted from the matched filter output of the PDRF,and the time-frequency joint feature is constructed.Based on the time-frequency joint feature,the naive Bayesian classifier(NBC)with minimal risk is established for target and jamming recognition.To improve the adaptability of the proposed method in complex environments,an online update process that adaptively modifies the classifier in the duration of the work of the PDRF is proposed.The experiments show that the PDRF can maintain high recognition accuracy when the signal-to-noise ratio(SNR)decreases and the jamming-to-signal ratio(JSR)increases.Moreover,the applicable analysis shows that he ONBCMR method has low computational complexity and can fully meet the real-time requirements of PDRF.

Deep Feature Bayesian Classifier for SAR Target Recognition with Small Training Set

In recent years,deep learning algorithms have been popular in recognizing targets in synthetic aperture radar(SAR)images.However,due to the problem of overfitting,the performance of these models tends to worsen when just a small number of training data are available.In order to solve the problems of overfitting and an unsatisfied performance of the network model in the small sample remote sensing image target recognition,in this paper,we uses a deep residual network to autonomously acquire image features and proposes the Deep Feature Bayesian Classifier model(RBnet)for SAR image target recognition.In the RBnet,a Bayesian classifier is used to improve the effect of SAR image target recognition and improve the accuracy when the training data is limited.The experimental results on MSTAR dataset show that the RBnet can fully exploit effective information in limited samples and recognize the target of the SAR images more accurately.Compared with other state-of-the-art methods,our method offers significant recognition accuracy improvements under limited training data.Noted that theRBnet is moderately difficult to implement and has the value of popularization and application in engineering application scenarios in the field of small-sample remote sensing target recognition and recognition.

Decision Bayes Criteria for Optimal Classifier Based on Probabilistic Measures

This paper addresses the high dimension sample problem in discriminate analysis under nonparametric and supervised assumptions. Since there is a kind of equivalence between the probabilistic dependence measure and the Bayes classification error probability, we propose to use an iterative algorithm to optimize the dimension reduction for classification with a probabilistic approach to achieve the Bayes classifier. The estimated probabilities of different errors encountered along the different phases of the system are realized by the Kernel estimate which is adjusted in a means of the smoothing parameter. Experiment results suggest that the proposed approach performs well.

Multilocus Phylogeny of Lycodon and the Taxonomic Revision of Oligodon multizonatum

Classification of the Asian snake genera Lycodon and Oligodon has proven challenging. We conducted a molecular phylogenetic analysis to estimate the phylogenetic relationships in the genus of Lycodon and clarify the taxonomic status of Oligodon multizonatum using mitochondrial（cyt b, ND4） and nuclear（c-mos） genes. Phylogenetic trees estimated using Maximum Likelihood and Bayesian Inference indicated that O. multizonatum is actually a species of Lycodon. Comparing morphological data from O. multizonatum and its closest relatives also supported this conclusion. Our results imply that a thorough review of the evolutionary relationships in the genus of Lycodon is strong suggested.

Robust Magnification Independent Colon Biopsy Grading System over Multiple Data Sources

Automated grading of colon biopsy images across all magnifications is challenging because of tailored segmentation and dependent features on each magnification.This work presents a novel approach of robust magnification-independent colon cancer grading framework to distinguish colon biopsy images into four classes:normal,well,moderate,and poor.The contribution of this research is to develop a magnification invariant hybrid feature set comprising cartoon feature,Gabor wavelet,wavelet moments,HSV histogram,color auto-correlogram,color moments,and morphological features that can be used to characterize different grades.Besides,the classifier is modeled as a multiclass structure with six binary class Bayesian optimized random forest(BO-RF)classifiers.This study uses four datasets(two collected from Indian hospitals—Ishita Pathology Center(IPC)of 4X,10X,and 40X and Aster Medcity(AMC)of 10X,20X,and 40X—two benchmark datasets—gland segmentation(GlaS)of 20X and IMEDIATREAT of 10X)comprising multiple microscope magnifications.Experimental results demonstrate that the proposed method outperforms the other methods used for colon cancer grading in terms of accuracy(97.25%-IPC,94.40%-AMC,97.58%-GlaS,99.16%-Imediatreat),sensitivity(0.9725-IPC,0.9440-AMC,0.9807-GlaS,0.9923-Imediatreat),specificity(0.9908-IPC,0.9813-AMC,0.9907-GlaS,0.9971-Imediatreat)and F-score(0.9725-IPC,0.9441-AMC,0.9780-GlaS,0.9923-Imediatreat).The generalizability of the model to any magnified input image is validated by training in one dataset and testing in another dataset,highlighting strong concordance in multiclass classification and evidencing its effective use in the first level of automatic biopsy grading and second opinion.

A Study of Electromyogram Based on Human-Computer Interface

In this paper, a new control system based on forearm electromyogram （EMG） is proposed for computer peripheral control and artificial prosthesis control. This control system intends to realize the commands of six pre-defined hand poses： up, down, left, right, yes, and no. In order to research the possibility of using a unified amplifier for both electroencephalogram （EEG） and EMG, the surface forearm EMG data is acquired by a 4-channel EEG measurement system. The Bayesian classifier is used to classify the power spectral density （PSD） of the signal. The experiment result verifies that this control system can supply a high command recognition rate （average 48%） even the EMG data is collected with an EEG system just with single electrode measurement.

Faulty Feeder Identification and Fault Area Localization in Resonant Grounding System Based on Wavelet Packet and Bayesian Classifier

Accurate fault area localization is a challenging problem in resonant grounding systems(RGSs).Accordingly,this paper proposes a novel two-stage localization method for single-phase earth faults in RGSs.Firstly,a faulty feeder identification algorithm based on a Bayesian classifier is proposed.Three characteristic parameters of the RGS(the energy ratio,impedance factor,and energy spectrum entropy)are calculated based on the zero-sequence current(ZSC)of each feeder using wavelet packet transformations.Then,the values of three parameters are sent to a pre-trained Bayesian classifier to recognize the exact fault mode.With this result,the faulty feeder can be finally identified.To find the exact fault area on the faulty feeder,a localization method based on the similarity comparison of dominant frequency-band waveforms is proposed in an RGS equipped with feeder terminal units(FTUs).The FTUs can provide the information on the ZSC at their locations.Through wavelet-packet transformation,ZSC dominant frequency-band waveforms can be obtained at all FTU points.Similarities of the waveforms of characteristics at all FTU points are calculated and compared.The neighboring FTU points with the maximum diversity are the faulty sections finally determined.The proposed method exhibits higher accuracy in both faulty feeder identification and fault area localization compared to the previous methods.Finally,the effectiveness of the proposed method is validated by comparing simulation and experimental results.

ASSERT:attack synthesis and separation with entropy redistribution towards predictive cyber defense

The sophistication of cyberattacks penetrating into enterprise networks has called for predictive defense beyond intrusion detection,where different attack strategies can be analyzed and used to anticipate next malicious actions,especially the unusual ones.Unfortunately,traditional predictive analytics or machine learning techniques that require training data of known attack strategies are not practical,given the scarcity of representative data and the evolving nature of cyberattacks.This paper describes the design and evaluation of a novel automated system,ASSERT,which continuously synthesizes and separates cyberattack behavior models to enable better prediction of future actions.It takes streaming malicious event evidences as inputs,abstracts them to edge-based behavior aggregates,and associates the edges to attack models,where each represents a unique and collective attack behavior.It follows a dynamic Bayesian-based model generation approach to determine when a new attack behavior is present,and creates new attack models by maximizing a cluster validity index.ASSERT generates empirical attack models by separating evidences and use the generated models to predict unseen future incidents.It continuously evaluates the quality of the model separation and triggers a re-clustering process when needed.Through the use of 2017 National Collegiate Penetration Testing Competition data,this work demonstrates the effectiveness of ASSERT in terms of the quality of the generated empirical models and the predictability of future actions using the models.

Classification of the firmness of peaches by sensor fusion

The objectives of this research were to compare the performance of each individual nondestructive sensor with the destructive sensor,and to apply sensor fusion technique to explore whether a combination of sensors would give better results than a single sensor for classification of peach firmness.Tests were carried out with four peach varieties namely Royal Glory,Caterina,Tirrenia and Suidring.In this research,the three nondestructive firmness sensors acoustic firmness,low-mass impact and micro-deformation impact were used to measure firmness.A Bayesian classifier was chosen to provide a classification into three categories,namely soft,intermediate and hard.High level fusion technique was performed by using identity declaration provided by each sensor.The data fusion system processed the information of the sensors to output the fused data.The result of the high level fusion was compared with the classification provided by an unsupervised algorithm based on destructive reference measurement.The fusion process of the nondestructive sensors provided some improvements in the firmness classification;the error rate varied from 25%to 19%for individual sensor.Furthermore,the results of fusion process by using three sensors decreased the error rate from 19%to 13%.This research demonstrated that the fused systems provided more complete and complementary information and,thus,were more effective than individual sensors in the firmness classification of peaches.

ASSERT:attack synthesis and separation with entropy redistribution towards predictive cyber defense

The sophistication of cyberattacks penetrating into enterprise networks has called for predictive defense beyond intrusion detection,where different attack strategies can be analyzed and used to anticipate next malicious actions,especially the unusual ones.Unfortunately,traditional predictive analytics or machine learning techniques that require training data of known attack strategies are not practical,given the scarcity of representative data and the evolving nature of cyberattacks.This paper describes the design and evaluation of a novel automated system,ASSERT,which continuously synthesizes and separates cyberattack behavior models to enable better prediction of future actions.It takes streaming malicious event evidences as inputs,abstracts them to edge-based behavior aggregates,and associates the edges to attack models,where each represents a unique and collective attack behavior.It follows a dynamic Bayesian-based model generation approach to determine when a new attack behavior is present,and creates new attack models by maximizing a cluster validity index.ASSERT generates empirical attack models by separating evidences and use the generated models to predict unseen future incidents.It continuously evaluates the quality of the model separation and triggers a re-clustering process when needed.Through the use of 2017 National Collegiate Penetration Testing Competition data,this work demonstrates the effectiveness of ASSERT in terms of the quality of the generated empirical models and the predictability of future actions using the models.