Search

Document forgeries that involve modification of the materials used, such as ink and paper, provide evidence of any malpractices being performed. Forensic specialists use different techniques to identify and classify these samples; however, the most preferred method is to use nondestructive techniques to avoid any potential damage to the original specimen under investigation. Hyperspectral imaging has already been explored in several application domains and used as a powerful method in forensic investigations to extract information about the materials under examination. To precisely classify the material information and utilize the hyperspectral imaging technique's potential, we probed the potential of some hybrid spectral similarity measures to classify different commonly used paper samples. A comparison of these methods is quantitatively presented in this article. Hybrid spectral similarity algorithms are tested on forensic analysis of paper data. We compared the classification capabilities of various hybrid spectral similarity algorithms on hyperspectral data of 40 different paper samples. The overall accuracy (OA), kappa K̂, Z‐score of kappa (ZK̂), and the 95% confidence interval of kappa (CI(K̂)) are used for comparison. The SID‐SAM and SID‐SCA produced an overall accuracy of 88% and 87%, respectively, which is highest among the hybrid spectral similarity measures tested. [ABSTRACT FROM AUTHOR]

Class noise is a common issue that affects the performance of classification techniques on real-world data sets. Class noise appears when a class variable in data sets has incorrect class labels. In the case of noisy data, the robustness of classification techniques against noise could be more important than the performance results on noise-free data sets. The decision tree method is one of the most popular techniques for classification tasks. The C4.5, CART, and random forest (RF) algorithms are considered to be three of the most used algorithms in decision trees. The aim of this paper is to reach conclusions on which decision tree algorithm is better to use for building decision trees in terms of its performance and robustness against class noise. In order to achieve this aim, we study and compare the performance of the models when applied to class variables with noise. The results obtained indicate that the RF algorithm is more robust to data sets with noisy class variable than other algorithms. [ABSTRACT FROM AUTHOR]

The naive Bayes classification algorithm is a simple and practical method for classification. There were a lot of studieson the assumption of conditional independence between attributes. The researchwas committed to eliminate redundant attributes and to reduce the correlation between attributes, with the aim to obtain some new attributes beingmore independentof and adapted to naive Bayes algorithm. However, the independence between new attributes was not easy to measure. Therefore, the improvement of the naive Bayes algorithm was not supported sufficiently by theory but was more supported by data experiments. This paper definedGauss distributed data and proposedan improved naive Bayes algorithm using the Gram-Schmidt orthogonalization method, making it convenient for classification with Gaussian distribution data. The improved method was different from the previous method. Unlike previous methods that explicitly construct new attributes set or attributes transformation matrix, the approach in the current paper directly orthogonalized the sample data of attributes. It was also provedthat the abstract new attributes corresponding to the orthogonalized attributes datawere independence. This showedthat the assumption of conditional independence in the original naive Bayes algorithm will not cause obstacles to the use of the algorithm for the classification of Gauss distributed data, as this constraint can be satisfied after Gram-Schmidt orthogonalization. [ABSTRACT FROM AUTHOR]

This paper presents a method for unsupervised classification of entities by a group of agents with unknown domains and levels of expertise. In contrast to the existing methods based on majority voting ("wisdom of the crowd") and their extensions by expectation-maximization procedures, the suggested method first determines the levels of the agents' expertise and then weights their opinions by their expertise level. In particular, we assume that agents will have relatively closer classifications in their field of expertise. Therefore, the expert agents are recognized by using a weighted Hamming distance between their classifications, and then the final classification of the group is determined from the agents' classifications by expectation-maximization techniques, with preference to the recognized experts. The algorithm was verified and tested on simulated and real-world datasets and benchmarked against known existing algorithms. We show that such a method reduces incorrect classifications and effectively solves the problem of unsupervised collaborative classification under uncertainty, while outperforming other known methods. [ABSTRACT FROM AUTHOR]

In the contemporary world, data pervades every facet of human life, and the information contained in this data plays a pivotal role in shaping decision-making and advancing technology. Among the plethora of techniques available, classification methods are highly effective tools for extracting valuable insights from vast volumes of data. The R ule A ggregation C lassifi ER (RACER) is a novel rule-based classification algorithm known for its exceptional performance. A notable limitation of RACER lies in its inability to handle continuous features. In this paper, we address the aforementioned limitation by employing various supervised discretization methods, including CAIM, MDLP, Decision Tree (CART), and ChiMerge. The impact of these methods on RACER's accuracy and understandability is evaluated across nine datasets from the UCI repository. Additionally, the paper conducts a comparative analysis of RACER's accuracy against well-known classifiers such as Naive Bayes, Logistic Regression, SVM, LightGBM, and Decision Tree. The findings indicate that RACER achieves the highest average accuracy when we utilize MDLP as the discretization method, surpassing its initial average accuracy. Moreover, RACER demonstrates superior understandability by generating the lowest number of rules when employing ChiMerge and Decision Tree for discretizing numerical features. Furthermore, RACER outperforms the other five classifiers when employing MDLP. • The discretization unit enables the RACER algorithm to handle continuous data. • The discretization unit improves the classification accuracy of the RACER. • The discretization unit increases the understandability of the RACER rules. [ABSTRACT FROM AUTHOR]

Most of the existing multi-label learning techniques consider the correlation learning problem but ignore the inconsistency feature of structural properties, which can change the structural property of the label space obtained by function mapping and it can directly affect the classification performance. In order to deal with this problem, this paper proposed a multi-label classification algorithm based on structural property preserving policy and correlation learning policies. Firstly, the algorithm constructed a mapping function to realize the mapping between the feature space and the label space. Secondly, it used structural property preserving strategy based on feature data to reduce the difference between feature data caused by linear transformation. Finally, this paper introduced the correlation learning strategy based on label data to further optimize the parameters of the proposed algorithm to improve the classification performance. The experimental results through a group of benchmark instances show that compared with the other classical multi-label classification algorithm, the proposed algorithm is much superior in classification performance, and the algorithm has great effectiveness. [ABSTRACT FROM AUTHOR]

Background: HER2 gene expression is one of the main reference indicators for breast cancer detection and treatment, and it is also an important target for tumor targeted therapy drug selection. Therefore, the correct detection and evaluation of HER2 gene expression has important value for clinical treatment of breast cancer.Objective: The study goal is to better classify HER2 images.Methods: For general convolution neural network, with the increase of network layers, over fitting phenomenon is often very serious, which requires setting the value of random descent ratio, and parameter adjustment is often time-consuming and laborious, so this paper uses residual network, with the increase of network layer, the accuracy will not be reduced.Results: In this paper, a HER2 image classification algorithm based on improved residual network is proposed. Experimental results show that the proposed HER2 network has high accuracy in breast cancer assessment.Conclusion: Taking HER2 images in Stanford University database as experimental data, the accuracy of HER2 image automatic classification is improved through experiments. This method will help to reduce the detection intensity and improve the accuracy of HER2 image classification. [ABSTRACT FROM AUTHOR]

Item Response Theory (IRT) has been proposed within the field of Educational Psychometrics to assess student ability as well as test question difficulty and discrimination power. More recently, IRT has been applied to evaluate machine learning algorithm performance on a single classification dataset, where the student is now an algorithm, and the test question is an observation to be classified by the algorithm. In this paper we present a modified IRT-based framework for evaluating a portfolio of algorithms across a repository of datasets, while simultaneously eliciting a richer suite of characteristics - such as algorithm consistency and anomalousness - that describe important aspects of algorithm performance. These characteristics arise from a novel inversion and reinterpretation of the traditional IRT model without requiring additional dataset feature computations. We test this framework on algorithm portfolios for a wide range of applications, demonstrating the broad applicability of this method as an insightful algorithm evaluation tool. Furthermore, the explainable nature of IRT parameters yield an increased understanding of algorithm portfolios. [ABSTRACT FROM AUTHOR]

Using feature selection technology in data mining to define the key features from high-dimensional and complex omics data has been the focus of research. This paper studied the complex correlation between features of the omics data, and proposed a feature module selection algorithm based on the synergetic network. The interaction gain was used to construct the weighted synergetic network. By measuring the closeness of candidate nodes to the current feature module and combining with the node’s own classification performance, important feature module can be obtained. The algorithm was validated on ten public data sets and showed advantages over the comparison algorithms in classification accuracy, sensitivity and specificity, which indicates the feature modules selected by the algorithm have better performance. [ABSTRACT FROM AUTHOR]

Subspace embedding is a popular technique to discover a mapping space in which the samples are expected to be represented appropriately. In recent years, graph has received increasing attention in subspace embedding and most of these related graph-based algorithms directly construct the connecting graph in original space. But some redundant information probably exists in the data with high dimension, and thus, it is hard to ensure the quality of graph. In this paper, we propose a novel discriminative subspace embedding (DSE) algorithm for classification. DSE is a supervised subspace learning method. In DSE, an intra-class graph and an inter-class graph are used to characterize the relationship among samples from the same class and different classes, respectively. DSE assumes that the embeddings of samples from the same class should be similar while different embeddings should be learned for the samples belonging to different classes. Based on this assumption, the above two graphs are constructed in mapping space. In order to enhance the quality of projections, the reconstruction of original data is also taken into consideration in DSE. Finally, some datasets are adopted to test the performance of DSE. Experimental results illustrate that effective representations can be learned by DSE and it has a more competitive learning ability, in comparison with related algorithms. [ABSTRACT FROM AUTHOR]

With the rapid development of the Internet, image information is explosively growing. Traditional image classification methods are difficult to deal with huge image data and cannot meet people's requirements on the accuracy and speed of image classification. In recent years, the convolutional neural network (CNN) has been developing rapidly, and it has performed extremely well. The image classification method based on CNN breaks through the bottleneck of traditional image classification methods and becomes the mainstream image classification algorithm at present. CT image classification algorithm is one of the research hot spots in the field of medical image. The purpose of this paper is to apply convolutional neural network to CT image classification, so as to speed up CT image classification and improve the accuracy of CT image classification and so as to reduce the workload of doctors and improve work efficiency. In this paper, CT images are classified by CDBN model. Vector machine SVM is used as the feature classifier of CDBN model to enhance feature transfer and reuse so as to enrich the features. It also suppresses features that are not very useful for current tasks and improves the performance of the model. Using CDBN to classify CT images, several commonly used gray images are compared. Comparing the results of the ordinary gradient algorithm with Adam algorithm, we can get the CDBN model using Adam optimization algorithm. In CT image classification, both accuracy and speed have a good effect. The experimental results show that the training speed of CDBN model of Adam optimization algorithm in CT image classification is 3% faster than that of general gradient algorithm. [ABSTRACT FROM AUTHOR]

Nowadays, imbalanced data classification is a hot topic in data mining and recently, several valuable researches have been conducted to overcome certain difficulties in the field. Moreover, those approaches, which are based on ensemble classifiers, have achieved reasonable results. Despite the success of these works, there are still many unsolved issues such as disregarding the importance of samples in balancing, determination of proper number of classifiers and optimizing weights of base classifiers in voting stage of ensemble methods. This paper intends to find an admissible solution for these challenges. The solution suggested in this paper applies the support vector data descriptor (SVDD) for sampling both minority and majority classes. After determining the optimal number of base classifiers, the selected samples are utilized to adjust base classifiers. Finally, genetic algorithm optimization is used in order to find the optimum weights of each base classifier in the voting stage. The proposed method is compared with some existing algorithms. The results of experiments confirm its effectiveness. [ABSTRACT FROM AUTHOR]

The effective classification of cloud manufacturing resources is a prerequisite for resource service packaging and subsequent manufacturing resource service retrieval. And it aims for the classification of high-dimensional features and multicategory processing equipment resources. At first this paper extracted resource attribute characteristic values and performed vectorization processing by XML documents which described the characte ristics of cloud manufacturing resource attributes . Then, based on the SVM algorithm, it established an automatic classification and prediction algorithm model for manufacturing resources. In order to improve the classification efficiency of the SVM classification algorithm, this paper introduced the SMO algorithm, and it used the grid search to find the optimal parameters of the SVM algorithm for optimizing the effect of the SVM algorithm. Finally, based on the original equipment information of a certain equipment resource set, the paper conducted a classification experiment on different processing equipment to verify the validity and feasibility of the classification model. [ABSTRACT FROM AUTHOR]

In this paper, a new optimized algorithm is presented to classify the liquids with high-level accuracy using raw data which show the transmission parameter (S 21 ) of liquids. The free space measurement method, one of the microwave spectroscopy methods, has been chosen to collect the S 21 parameter for use in detecting illegal and explosive liquids. The aim is to classify the liquids that the passenger can choose with his/her plane journey. To classify and identify these liquids, instead of a multivariate data analysis method, a simpler dynamic time warping algorithm method is proposed. It is advantageous to intervene in this algorithm to perform the classification process better. It will be able to determine the most contradictory similarities among the signals collected from liquids and perform the best classification process. Thanks to this feature, it is expected to give better results than the algorithms proposed earlier. Because, it is thought that the liquids can be classified by the S 21 parameter which has unique properties of liquids. Furthermore, this algorithm will allow an intermediate group of suspicious liquids to be detected in another group. The performance of the proposed model is discussed in terms of its potential for reducing the classification problem and composing a database which can be used to identify the un/known materials easily. [ABSTRACT FROM AUTHOR]

The development of society has brought countless data, with the unbalancedness becoming a significant feature of many data sets. So it has come to be a research hotspot for machine learning on how to make those unbalanced data sets obtain better effects of classification. Based on this, this paper conducted a comprehensive research on the current unbalanced data set classification method, and made an overall interpretation and conclusion from such three aspects as the unbalanced data sampling method, the method of machine learning-based improved algorithm and the combination method. It also analyzed and took int o account many factors, including the problems solved by each method, algorithm mentality, application scenarios, as well as the advantages and disadvantages of each, and delivered a summary on potential problems of the classification methods and a prospect on the future research directions [ABSTRACT FROM AUTHOR]

Dimensionality reduction based on random projection (RP) includes two problems, namely, the dimensionality is limited by the data size and the class separability of the dimensionality reduction results is unstable due to the randomly generated projection matrix. These problems make the RP algorithm unsuitable for large-size hyperspectral image (HSI) classification. To solve these problems, this paper presents a new partitioned RP (PRP) algorithm and proves its rationality in theory. First, a large-size HSI is evenly divided into multiple small-size sub-HSIs. Afterwards, the projection matrix that maximizes the class separability is selected from multiple samplings in which the class dissimilarity measurement is defined as large inter-class distance and small intra-class variance. By using the same projection matrix, each small-size sub-HSI is projected to generate a low dimensional sub-HSI, thereby generating a low dimensional HSI. Next, the minimum distance (MD) classifier is utilized to classify the low dimensional HSI obtained by the PRP algorithm. Finally, four real HSIs are used for experiments, and three of the most popular classification algorithms based on RP are selected as comparison algorithms to validate the effectiveness of the proposed algorithm. The classification performance is evaluated with the kappa coefficient, overall accuracy (OA), average accuracy (AA), average precision rate (APR), and running time. Experimental results indicate that the proposed algorithm can obtain reliable classification results in a very short time. [ABSTRACT FROM AUTHOR]

In supervised classification, obtaining nonlinear separating functions from an algorithm is crucial for prediction accuracy. This paper analyzes the polyhedral conic functions (PCF) algorithm that generates nonlinear separating functions by only solving simple subproblems. Then, a revised version of the algorithm is developed that achieves better generalization and fast training while maintaining the simplicity and high prediction accuracy of the original PCF algorithm. This is accomplished by making the following modifications to the subproblem: extension of the objective function with a regularization term, relaxation of a hard constraint set and introduction of a new error term. Experimental results show that the modifications provide %12 better generalization on average and up to 10x faster training. This paper also contributes to the literature by providing detailed comparisons of the other classification algorithms that use polyhedral conic functions for the first time. [ABSTRACT FROM AUTHOR]

In this paper, we propose a multicategory large margin classification with unequal costs. In addition to extending the standard multicategory SVM to the case of unequal costs, we also develop an unequal costs classification for ψ-loss and propose an efficient algorithm for computation. Besides commonly used L2 penalty, the adaptive LASSO is also examined to remove irrelevant variables. Theoretically, we derive the Bayes rules under the generalized cost and reduce the infinite sum-to-zero constraint to a finite constraint. Numerically, we demonstrate the good performance of our methodology on two simulated examples and one real-life dataset. [ABSTRACT FROM AUTHOR]

The traditional naive Bayes classification algorithm does not divide the importance degree according to the different feature items, which makes the classification result inaccurate. In order to solve this problem, this paper introduced Jensen-Shannon(JS) divergence and used JS divergence to express the amount of information provided by the feature terms. Aiming at the deficiency of JS divergence, the paper considered from the three aspects of word frequency, text frequency and inverse category frequency corrected by coefficient of variation, and then adjusted and corrected the JS divergence. Finally, it calculated the weight of each feature and introduced the weights into the naive Bayes formula. Compared with other algorithms, it proves that this method improves the naive Bayes classification algorithm effectively. Therefore, compared with other classification algorithms, the performance of naive Bayes classification algorithm based on JS divergence feature weighting is greatly improved. [ABSTRACT FROM AUTHOR]

Student academic accomplishment is the foremost focus of every educational institution. In developing student achievement in educational institutions, the researchers finally created a new research area, namely educational data mining (EDM). How the feature selection (FS) algorithm works is by removing unrelated data from educational datasets; therefore, this algorithm can improve the classification performance managed in EDM techniques. This research presents an analysis of the performance of the FS algorithm from the student dataset. The results received from other FS algorithms and classifiers will help other researchers to gain some best combination regarding FS algorithms and the classification. Selecting features that are relevant for student forecast models is a sensitive problem to stakeholders in education because they must make decisions based on the results of the prediction models. For the future, our paper seeks to play a decisive part while developing quality concerning education, as well as guiding different researchers in conducting educational interventions. [ABSTRACT FROM AUTHOR]

The recognition of hand movements is an important method for human-computer interaction (HCI) technology, and it is widely used in virtual reality and other HCI areas. While many valuable efforts have been made, efficient ways to capture over 20 types of hand movements with high accuracy by one data glove are still lacking. This paper addresses a new classification framework for 52 hand movements. This classification framework includes the following two parts: the movement detection algorithm and the movement classification algorithm. The fine K-nearest neighbor (Fine KNN) is the core of the movement detection algorithm. The movement classification algorithm is composed of downsampling in data preparation and a new deep learning network named the DBDF network. Bidirectional Long Short-Term Memory (BiLSTM) is the main part of the DBDF network. The results of experiments using the Ninapro DB1 dataset demonstrate that our work can classify more types of hand movements than related algorithms with a precision of 93.15%. [ABSTRACT FROM AUTHOR]

Least squares twin support vector machine (LSTSVM) is an effective and efficient learning algorithm for pattern classification. However, the distance in LSTSVM is measured by squared L 2 -norm metric that may magnify the influence of outliers. In this paper, a novel robust least squares twin support vector machine framework is proposed for binary classification, termed as C L 2 , p -LSTSVM, which utilizes capped L 2 , p -norm distance metric to reduce the influence of noise and outliers. The goal of C L 2 , p -LSTSVM is to minimize the capped L 2 , p -norm intra-class distance dispersion, and eliminate the influence of outliers during training process, where the value of the metric is controlled by the capped parameter, which can ensure better robustness. The proposed metric includes and extends the traditional metrics by setting appropriate values of p and capped parameter. This strategy not only retains the advantages of LSTSVM, but also improves the robustness in solving a binary classification problem with outliers. However, the nonconvexity of metric makes it difficult to optimize. We design an effective iterative algorithm to solve the C L 2 , p -LSTSVM. In each iteration, two systems of linear equations are solved. Simultaneously, we present some insightful analyses on the computational complexity and convergence of algorithm. Moreover, we extend the C L 2 , p -LSTSVM to nonlinear classifier and semi-supervised classification. Experiments are conducted on artificial datasets, UCI benchmark datasets, and image datasets to evaluate our method. Under different noise settings and different evaluation criteria, the experiment results show that the C L 2 , p -LSTSVM has better robustness than state-of-the-art approaches in most cases, which demonstrates the feasibility and effectiveness of the proposed method. [ABSTRACT FROM AUTHOR]

The high-dimensional features in the data may affect the performance of the classification model as all of them are not useful. The selection of relevant optimal features is a tedious task, especially the data in which the number of features is high. This paper proposed a new feature selection (FS) approach based on the artificial electric field algorithm (AEFA) called FSAEFA, to select the most suitable optimal features for classification. The AEFA is an efficient and effective population-based optimization technique inspired by the principle of Coulomb's law of electrostatic force (CLEF) to solve optimization problems. The proposed FSAEFA approach can search the most suited subset of features by attracting the worst individual features towards the best individual features by using the attraction of CLEF. The proposed FSAEFA approach has been evaluated and compared with some other FS approaches on ten publicly available benchmark datasets. The experimental result has been indicated that the performance of the proposed method is superior over the existing methods in most of the cases. The significance of the proposed FS approach along with their counter parts have been statistically measured and compared by using Friedman test and Holm procedure. It has been determined that the proposed FSAEFA approach is found to be more efficient than the existing FS approaches. [ABSTRACT FROM AUTHOR]

In classification, a decision tree is a common model due to its simple structure and easy understanding. Most of decision tree algorithms assume all instances in a dataset have the same degree of confidence, so they use the same generation and pruning strategies for all training instances. In fact, the instances with greater degree of confidence are more useful than the ones with lower degree of confidence in the same dataset. Therefore, the instances should be treated discriminately according to their corresponding confidence degrees when training classifiers. In this paper, we investigate the impact and significance of degree of confidence of instances on the classification performance of decision tree algorithms, taking the classification and regression tree (CART) algorithm as an example. First, the degree of confidence of instances is quantified from a statistical perspective. Then, a developed CART algorithm named C_CART is proposed by introducing the confidence of instances into the generation and pruning processes of CART algorithm. Finally, we conduct experiments to evaluate the performance of C_CART algorithm. The experimental results show that our C_CART algorithm can significantly improve the generalization performance as well as avoiding the over-fitting problem to a certain extend. [ABSTRACT FROM AUTHOR]

Deep Packet Inspection (DPI) of the network traffic is used on a regular basis within the traditional and virtualized environments. But changes in the network architecture with the introduction of containers, microservices, application functions, network functions, and the penetration of 5G access technology are adding more traffic complexity, especially in the so-called eastwest flow direction. Network Functions Virtualization (NFV) has become an unavoidable step for further IP network development. In this context, DPI is becoming a challenge. Furthermore, the penetration of 5G allows access of various kinds of devices to the network with cloudification logic which drives them. This paper provides a performance analysis of a selected set of supervised machine learning (ML) algorithms for classification of network traffic within an NFV environment. The goal is to find a suitable algorithm that will classify the traffic from a point of both precision and speed, especially because in the 5G networks any packet delay may compromise the quality of service requirements. The research shows that out of the 6 algorithms tested, Decision Tree algorithm has the best overall performance, from both classification precision and time consumption point of view. It has proved as a reliable classifier that is performing evenly across different classes. Due to the specifics of the virtualized environments and encryption methods, payload data, source, destination, and port information of the network traffic packets are excluded from any statistical operation used for classification by the ML algorithms. [ABSTRACT FROM AUTHOR]

In the bioinformatics field, the classification of unknown biological sequences is a key task that is fundamental for simplifying the consistency, aggregation, and survey of organisms and their evolution. We can view biological sequences as data components of higher non-fixed dimensions, corresponding to the length of the sequences. Numerical encoding performs an important function in DNA sequence evaluation via computational procedures such as one-hot encoding (OHE). However, the OHE method has drawbacks: 1) it does not add any details that may produce the additional predictive variable, and 2) if the variable has many classes, then OHE increases the feature space significantly. To overcome these drawbacks, this paper presents a computationally effective framework for classifying DNA sequences of living organisms in the image domain. The proposed strategy relies upon multilayer perceptron trained by a pseudoinverse learning autoencoder (PILAE) algorithm. The PILAE training process does not have to set the learning control parameters or indicate the number of hidden layers. Therefore, the PILAE classifier can accomplish better performance contrasting with other deep neural network (DNNs) strategies such as VGG-16 and Xception models. Experimental results have demonstrated that this proposed strategy achieves high prediction accuracy as well as to a significant degree high computational efficiency over different datasets. [ABSTRACT FROM AUTHOR]

[Display omitted] The unstable statuses in machining, such as chatter and severe forced vibration, limit manufacturing productivity and quality. Although there have been plentiful studies on the on-line sensing signals and off-line surface images to identify these statuses qualitatively, the quantification of chatter severity, which is of great significance on machining stability, has been rarely reported. Workpiece surface morphology is the direct result and thus the objective criteria of unstable machining. This paper proposes a new algorithm to quantify the chatter status based on the milled surface image analysis of end-milling. Besides, the proposed method can also identify the stable, forced vibration, and chatter statuses. Specifically, this proposed algorithm extracts two indexes, called Index_Dmean and Index_Emean , via processing the milled surface image. The definitions and calculations of these two indexes are given in Section 2. Index_Dmean is used to differentiate stable and unstable surfaces and quantify the severity of chatter status, and Index_Emean can distinguish between forced vibration and chatter surfaces. Experimental results demonstrate the availability of the proposed algorithm on the chatter quantification and milling instability classification performance. In addition, an applicable on-line vision system with satisfying performance based on the algorithm is developed. [ABSTRACT FROM AUTHOR]

Aiming at the problem of the discriminator' s poor classification accuracy on a small number of labeled samples and insufficient robustness to the local perturbation of manifolds in the generative adversarial network, this paper proposed a generative adversarial network based on variable loss and manifold regularization. The algorithm used a variable loss instead of the original discriminator to solve the adverse effect of the poorly trained classifier on the semi-supervised classification task. In addition, on the basis of variable loss of discriminator, it added manifold regular terms to improve the robustness of discriminator to local disturbance by punishing the variation of classification decision of discriminator on manifold. Using the quality of the generated samples and the semi-supervised classification accuracy as the evaluation criteria of the algorithm, it performed numerical experiments on the dataset SVHN and CIFAR-10. Comparing with other semi-supervised algorithms, the results show that the proposed algorithm can obtain higher quality generated samples and higher precision classification results with a small amount of labeled data. [ABSTRACT FROM AUTHOR]

In hyperspectral image (HSI) classification, a big challenge is the limited sample size with a relatively high feature dimension. Therefore, effective feature extraction of data is essential, which is desired to remove the redundancy as well as improve the discrimination. A huge number of methods have been proposed for HSI feature extraction. In recent years, deep learning-based feature extraction algorithms have shown their superiorities in various classification problems. Within them, deep PCA (DPCA) is a simple but efficient algorithm, which runs fast due to the absence of back-propagation. However, DPCA fails to provide satisfactory classification accuracies on HSI datasets. In this paper, we try to combine DPCA with residual-based multi-scale feature extraction and propose a residual deep PCA (RDPCA) feature extraction algorithm for HSI classification. It is a hierarchical approach consisting of multiple layers. Within each layer, PCA is utilized for layer-wise feature extraction, and the reconstruction residual is fed into the next layer. When the feature is passed deeper into the RDPCA network, finer details are mined. The layer-wise features are concatenated to form the final output feature. Furthermore, to enhance the ability of nonlinear feature extraction, we add activation functions between adjacent layers. Experimental results on real-world HSI datasets have shown the superiority of the proposed RDPCA over DPCA and PCA. [ABSTRACT FROM AUTHOR]

In order to overcome the problems of poor accuracy and high complexity of current classification algorithm for non-equilibrium data set, this paper proposes a decision tree classification algorithm for non-equilibrium data set based on random forest. Wavelet packet decomposition is used to denoise non-equilibrium data, and SNM algorithm and RFID are combined to remove redundant data from data sets. Based on the results of data processing, the non-equilibrium data sets are classified by random forest method. According to Bootstrap resampling method with certain constraints, the majority and minority samples of each sample subset are sampled, CART is used to train the data set, and a decision tree is constructed. Obtain the final classification results by voting on the CART decision tree classification. Experimental results show that the proposed algorithm has the characteristics of high classification accuracy and low complexity, and it is a feasible classification algorithm for non-equilibrium data set. [ABSTRACT FROM AUTHOR]

G-Protein-Coupled Receptors (GPCR) are the large family of protein membrane; and until now some of them still remain orphans. Predicting GPCR functions is a challenging task, it depends closely to their classification, which requires a digital representation of each protein chain as an attribute vector. A major problem of GPCR databases is their great number of features which can produce combinatorial explosion and increase the complexity of classification algorithms. Feature selection techniques are used to deal with this problem by minimizing features space dimension, and keeping the most relevant ones. In this paper, we propose to use the BAT algorithm for extracting the pertinent features and to improve the classification results. We compared the results obtained by our system with two other bio-inspired algorithms, Evolutionary Algorithm and PSO search. Metrics quality measures used for comparison are Error Rate, Accuracy, MCC and F -measure. Experimental results indicate that our system is more efficient. [ABSTRACT FROM AUTHOR]

Aiming at the problems of low recall rate and poor running efficiency caused by traditional abnormal accounts detecting algorithms in non-balanced social network datasets, the paper extracted user content, behavior, attributes, and relationship features from social network data sets, selected features using gradient-enhanced ensemble classifier XGBoost algorithm, established classification model,constructed unbalanced data sets and realized the identification of three types of spam accounts. Experimental results show that it improves the recall rate and the F1 value in identification of three types of abnormal users effectively by XGBoost algorithm in binary classification and multiple classification tasks both in the balanced and unbalanced dataset in comparison with the traditional classification methods such as random forest. And with few features selected by XGBoost, the classification algorithm can get the same effect as with all features of samples, which proves the effectiveness of this method. [ABSTRACT FROM AUTHOR]

In recent years, the sources localization has noticed an increase in research conducted on the problem of mixed far-field sources (FFSs) and near-field sources (NFSs). The main assumption of the existing researches is that the signals should be uncorrelated. Therefore, they cannot be used for multipath environments. The present paper provides a method called components separation algorithm (CSA) for the localization of multiple mixed FFSs and NFSs, including uncorrelated, lowly correlated and coherent signals. Firstly, by constructing one special cumulant matrix, and using a MUSIC-based technique, the noncoherent DOA vector (NDOAV) is extracted. By constructing another special cumulant matrix, and with respect to NDOAV, an estimate of the range, as well as a signal classification is obtained for noncoherent sources. Then, by estimating their kurtosis, the noncoherent component and consequently the coherent one of the second cumulant matrix is obtained. Finally, by introducing a novel approach based on squaring, projection, spatial smoothing, array interpolation transform and coherent component restoring, the parameters of coherent signals in each coherent group are estimated separately. The CSA prevents severe loss of the aperture. Furthermore, it does not require any pairing. The simulation results validate its satisfactory performance in terms of estimation accuracy, resolution, computational complexity, reasonable classification, and also its robustness against lowly correlated sources. [ABSTRACT FROM AUTHOR]

Classification algorithms' performance could be enhanced by selecting many representative points to be included in the training sample. In this paper, a new border and rare biased sampling (BRBS) scheme is proposed by assigning each point in the dataset an importance factor. The importance factor of border points and rare points (i.e. points belong to rare classes) is higher than other points. Then the points are selected to be in the training sample depending on these factors. Including these points in the training sample enhances classifiers experience. The results of experiments on 10 UCI machine learning repository datasets prove that the BRBS algorithm outperforms many sampling algorithms and enhanced the performance of several classification algorithms by about 8%. BRBS is proposed to be easy to configure, covering all points space, and generate a unique samples every time it is executed. [ABSTRACT FROM AUTHOR]

In order to improve the classification effect of the gene expression profile with high dimension and small sample, this paper proposes an improved rotation forest algorithm. Firstly, the information index to classification algorithm (IIC) can't deal with multi-class problems, and then an improved IIC algorithm is proposed to achieve the attribute reduction so that the noise genes are eliminated and the dimensions of feature space are reduced. Secondly, the rotation forest algorithm was improved from the two aspects of improving diversity and accuracy of the base classifiers to classify gene expression profiles. The simulation experimental results on the benchmark gene expression profile datasets show that our proposed algorithm is better than rotation forest algorithm in classification accuracy, stability and running time. [ABSTRACT FROM AUTHOR]

The rivality index (RI) is a normalized distance measurement between a molecule and their first nearest neighbours providing a robust prediction of the activity of a molecule based on the known activity of their nearest neighbours. Negative values of the RI describe molecules that would be correctly classified by a statistic algorithm and, vice versa, positive values of this index describe those molecules detected as outliers by the classification algorithms. In this paper, we have described a classification algorithm based on the RI and we have proposed four weighted schemes (kernels) for its calculation based on the measuring of different characteristics of the neighbourhood of molecules for each molecule of the dataset at established values of the threshold of neighbours. The results obtained have demonstrated that the proposed classification algorithm, based on the RI, generates more reliable and robust classification models than many of the more used and well-known machine learning algorithms. These results have been validated and corroborated by using 20 balanced and unbalanced benchmark datasets of different sizes and modelability. The classification models generated provide valuable information about the molecules of the dataset, the applicability domain of the models and the reliability of the predictions. [ABSTRACT FROM AUTHOR]

Deep learning breakthrough stimulates new research trends in civil infrastructure inspection, whereas the lack of quality-guaranteed, human-annotated, free-of-charge, and publicly available defect datasets with sufficient amounts of data hinders the progress of deep learning in defect inspection. To boost research in deep learning-based visual defect inspection, this paper first reviews and summarizes 40 publicly available defect datasets, covering common defects in various types of buildings and infrastructures. The taxonomy of the datasets is proposed based on specific deep learning objectives (classification, segmentation, and detection). Clarifications are also made for each dataset regarding its corresponding data volume, data resolution, data source, defect categories covered, infrastructure types focused, material types targeted, algorithms adopted for validation, annotation levels, context levels, and publication license for future utilization. Consequently, the summarized defect datasets offer around 13. 38 M labeled images, cover more than 5 defect types, 5 infrastructure types, 5 material types, and 3 levels of image context. Given that the crack is a common interest in civil engineering, this paper further combines existing datasets with self-labeled crack images to establish a benchmark dataset providing more than 15,000 and 11,000 labeled images for crack classification and segmentation, respectively. Based on the established crack dataset, experiments are conducted for classification, segmentation, and the subsequent non-maximum suppression-based detection tasks. The proposed multi-branch self-attention module and multi-stage-fused attentional pyramid network have been successfully adapted into the state-of-the-art (SOTA) classification network- Swin Transformer and segmentation networks including DeepLab V3+ , DenseNet , and Full Resolution ResNet. The resulting classification network achieves 88.0% accuracy, and the adapted segmentation models reach 77. 8 % , 77. 6 % , 76. 9 % mIoU (mean Intersection over Union), respectively. Moreover, a comprehensive comparison between 11 SOTA classification algorithms and 12 SOTA segmentation algorithms has been conducted. The algorithms proposed in this work are shown to achieve satisfactory performance with an acceptable efficiency on modern graphic processing units. Detailed suggestions are provided for constructing high-quality datasets and inspection algorithms. Finally, this paper remarks on the quantity, diversity, difficulty, and scalability of the reviewed defect datasets, feasibility on robotic platforms, superiority of proposed algorithms, and criticality of algorithm comparison results, formulating a solid baseline for future defect inspection research. • Review of the datasets for deep learning-based visual defect inspection. • Comparison of the algorithms for defect classification, segmentation and detection. • Proposed deep learning-based network architectures for defect inspection. • Suggestions on developing defect datasets and defect inspection algorithms. [ABSTRACT FROM AUTHOR]

Learning a classifier from imbalanced data is a challenging problem in Machine learning. A dataset is said to be imbalanced when the number of instances belonging to one class is much less than the number of instances belonging to the other class. Classifiers that proves efficient on standard data fail when the data is imbalanced as they are over trained by the majority class instances. Since class imbalance is a common characteristic of real-world data, the need for better classifiers becomes essential. This paper proposes a novel instance-based classification algorithm called Weighted Pattern Matching based Classification (PMC+) for classifying imbalanced data. PMC+ classifies unlabelled instances by computing the absolute difference between the feature values of the instances in the dataset and the unlabelled instance. PMC+ employs a simple classification procedure with weights and shows reasonably good performance. To improve the performance of PMC+, Fireworks based Feature and Weight Selection algorithm based on the idea of PMC+ has been proposed. PMC+ is evaluated on 44 binary imbalanced datasets and 15 multiclass imbalanced datasets. Although PMC+ does not employ a resampling or cost-sensitive method, experiments show that PMC+ is effective for classification of imbalanced data. The results of the experiments were validated using various non-parametric statistical tests. [ABSTRACT FROM AUTHOR]

The modelability index of a dataset of molecules is a measurement of the capacity of the dataset to be modeled using a QSAR algorithm. This measure allows to predict the correct classification rate of the dataset counting the nearest neighbors to the molecules of the dataset belonging to their same class. In this paper, we propose a new measure for the prediction of the modelability of datasets based on the use of the nearest neighbors based rivality index and the centroids based rivality index. These indexes take into account the noise that the nearest neighbor belonging to a different class could generate in the results of the QSAR classification algorithm. Using thirty benchmark datasets, two types of dataset representation and six different algorithms, we show the excellent behavior of the proposed indexes, obtaining correlations with values of R2 greater than 0.9 between the correct classification rate obtained in the classification processes using five folds cross-validation and the modelability index calculated using the centroid based rivality index. [ABSTRACT FROM AUTHOR]

Automatic color classification for solar cells is challenging because of the tiny color difference and low contrast. To address this problem, a color feature selection and classification frame is proposed in this paper. First, an intuitive multi-color space feature performance evaluation scheme is presented to select the optimal color subspaces that help to enormously enlarge the tiny color difference of solar cell images. And the optimal color subspaces can also be illustrated by employing multi-color space visualization method with combinational Mosaic images. Second, a nine-dimensional feature vector consisting of mean, variance, and skewness in the three optimal subspaces is extracted by utilizing the serial fusion technique. Third, an improved Gaussian mixture model in supervised manner for color classification is proposed by employing a k-means method based on adjacent rules, which helps to eliminate isolated points and enhances the classification performance. Finally, experimental results show that the overall performance of the proposed method achieves 97.9%, and outperform other algorithms especially for the tiny color difference of solar cell images. [ABSTRACT FROM AUTHOR]

This paper addresses the problem classification of instances/questions based on the opinions (classes) provided by anonymous agents. The solution aggregates the agents' classifications, aiming to obtain as close as possible to an unknown correct classification. However, the agents' fields or domains of competence and their levels of expertise are unknown and can vary extensively. Many popular classification algorithms address such a problem by following a "wisdom-of-the-crowd" approach while using different voting methods and expectation–maximization techniques. These algorithms lead to correct classifications when the majority of the agents are experts, thus classifying the instances correctly. However, they often result in erroneous classification when only a small subset of the agents are indeed correct. Moreover, these algorithms often assume a fixed set of classes for all instances. This study presents a fast (one-pass) classification algorithm that can estimate the unknown agents' expertise level and aggregates their classifications accordingly, even when these are obtained from different questionnaires; thus, when the instances are not necessarily classified to a fixed set of classes. The proposed algorithm finds the experts and the nonexpert agents for each question by analyzing the distance between them. The algorithm identifies the expert agents for each instance and then classifies them accordingly. The suggested algorithm is validated and compared against known methods by using both simulated datasets and real-world datasets collected from various sources. The obtained results clearly demonstrate the effectiveness and advantages of the proposed method. [ABSTRACT FROM AUTHOR]

The pull test (PT) is a common practice to assess the postural instability of patients with Parkinson’s disease. Postural instability is a serious issue for elderly and people with neurological disease, which can cause falls. The implementation of the PT consists in observing the user response after providing a tug to the patients’ shoulders, in order to displace the center of gravity from its neutral position. The validity of the test can be compromised by a nonstandard backward tug provided to the patient. The solution proposed in this paper consists of a low-cost multisensor system allowing an instrumented estimation of the input solicitation. Moreover, the system provides supplementary information on the user postural stability, by means of a set of features extracted from the user stabilogram. A wide set of experiments have been performed to assess the system capability to provide a rough classification between stable and unstable behaviors. Results obtained demonstrate the validity of the approach proposed, with very low rates of false positive and false negative. [ABSTRACT FROM PUBLISHER]

• Microalgae identification to solve the issues of harmful algae blooms. • The classification of potential algae strains for valuable bioactive compounds. • Machine learning and deep learning techniques for smart algae identification. • Image acquisition with consideration of robust dataset to enhance accuracy. • Incorporate various image pre-processing module to improve classification results. The identification of microalgae species is an important tool in scientific research and commercial application to prevent harmful algae blooms (HABs) and recognizing potential microalgae strains for the bioaccumulation of valuable bioactive ingredients. The aim of this study is to incorporate rapid, high-accuracy, reliable, low-cost, simple, and state-of-the-art identification methods. Thus, increasing the possibility for the development of potential recognition applications, that could identify toxic-producing and valuable microalgae strains. Recently, deep learning (DL) has brought the study of microalgae species identification to a much higher depth of efficiency and accuracy. In doing so, this review paper emphasizes the significance of microalgae identification, and various forms of machine learning algorithms for image classification, followed by image pre-processing techniques, feature extraction, and selection for further classification accuracy. Future prospects over the challenges and improvements of potential DL classification model development, application in microalgae recognition, and image capturing technologies are discussed accordingly. [ABSTRACT FROM AUTHOR]

This paper presents the intrinsic limit determination algorithm (ILD Algorithm), a novel technique to determine the best possible performance, measured in terms of the AUC (area under the ROC curve) and accuracy, that can be obtained from a specific dataset in a binary classification problem with categorical features regardless of the model used. This limit, namely, the Bayes error, is completely independent of any model used and describes an intrinsic property of the dataset. The ILD algorithm thus provides important information regarding the prediction limits of any binary classification algorithm when applied to the considered dataset. In this paper, the algorithm is described in detail, its entire mathematical framework is presented and the pseudocode is given to facilitate its implementation. Finally, an example with a real dataset is given. [ABSTRACT FROM AUTHOR]

One of the significant challenges in machine learning is the classification of imbalanced data. In many situations, standard classifiers cannot learn how to distinguish minority class examples from the others. Since many real problems are unbalanced, this problem has become very relevant and deeply studied today. This paper presents a new preprocessing method based on Delaunay tessellation and the preprocessing algorithm SMOTE (Synthetic Minority Over-sampling Technique), which we call DTO-SMOTE (Delaunay Tessellation Oversampling SMOTE). DTO-SMOTE constructs a mesh of simplices (in this paper, we use tetrahedrons) for creating synthetic examples. We compare results with five preprocessing algorithms (GEOMETRIC-SMOTE, SVM-SMOTE, SMOTE-BORDERLINE-1, SMOTE-BORDERLINE-2, and SMOTE), eight classification algorithms, and 61 binary-class data sets. For some classifiers, DTO-SMOTE has higher performance than others in terms of Area Under the ROC curve (AUC), Geometric Mean (GEO), and Generalized Index of Balanced Accuracy (IBA). [ABSTRACT FROM AUTHOR]

In this paper a model of the classification task, based on the Monte Carlo theory of stochastic algorithms, is presented. The model allows different classification paradigms to be described, such as the single classifier, the ensemble, and the crowdsourcing ones. Some experimental work validates the theoretical predictions about the relatioship between the mean and variance of the classification accuracy and the number of classifiers. [ABSTRACT FROM AUTHOR]

Abstract: Individuals from Pittsburgh rule-based classifiers represent a complete solution to the classification problem and each individual is a variable-length set of rules. Therefore, these systems usually demand a high level of computational resources and run-time, which increases as the complexity and the size of the data sets. It is known that this computational cost is mainly due to the recurring evaluation process of the rules and the individuals as rule sets. In this paper we propose a parallel evaluation model of rules and rule sets on GPUs based on the NVIDIA CUDA programming model which significantly allows reducing the run-time and speeding up the algorithm. The results obtained from the experimental study support the great efficiency and high performance of the GPU model, which is scalable to multiple GPU devices. The GPU model achieves a rule interpreter performance of up to 64 billion operations per second and the evaluation of the individuals is speeded up of up to 3.461×when compared to the CPU model. This provides a significant advantage of the GPU model, especially addressing large and complex problems within reasonable time, where the CPU run-time is not acceptable. [Copyright &y& Elsevier]

• Discretization is an important process in terms of both machine learning and data mining. • The effect of discretization on ensemble methods was analyzed. • Four Chi-square based discretization algorithms were used. • The Ensemble methods performed better on discrete data sets. Classification is one of the important tasks in data mining and machine learning. Classification performance depends on many factors as well as data characteristics. Some algorithms are known to work better with discrete data. In contrast, most real-world data contain continuous variables. For algorithms working with discrete data, these continuous variables must be converted to discrete ones. In this process called discretization, continuous variables are converted to their corresponding discrete variables. In this paper, four Chi-square based supervised discretization algorithms ChiMerge(ChiM), Chi2, Extended Chi2(ExtChi2) and Modified Chi2(ModChi2) were used. In the literature, the performance of these algorithms is often tested with decision trees and Naïve Bayes classifiers. In this study, differently, four sets of data discretized by these algorithms were classified with ensemble methods. Classification accuracies for these data sets were obtained through using a stratified 10-fold cross-validation method. The classification performance of the original and discrete data sets of the methods is presented comparatively. According to the results, the performance of the discrete data is more successful than the original data. [ABSTRACT FROM AUTHOR]

• We integrate a PSO with a NN and ERNN to form PSO-NN and PSO-ERNN metaheuristic models. • The PSO trains the NN and ERNN to achieve faster convergence and avoid local minima. • The PSO-NN and PSO-ERNN models are evaluated using ten benchmark classification datasets. • The accuracy and computational efficiency of the models are improved. • The PSO-ERNN model outperforms the PSO-NN model for most of the tested datasets. There are several types of neural networks (NNs) that are widely used for data classification tasks. The supervised learning NN is an advanced network with a training algorithm for setting the weights and biases of the network in its training phase. However, traditional training algorithms such as backpropagation have some drawbacks, such as slow convergence speed and falling into local minima, which reduces the performance of the classifier. Therefore, different nature-inspired metaheuristic algorithms are integrated with the NN training algorithms to provide derivative-free solutions for complex classification problems. Consequently, this paper proposes the integration of a particle swarm optimization (PSO) algorithm with an improved Elman recurrent neural network (ERNN) to form a PSO-ERNN metaheuristic model. The key contribution of this study is the development of a new dimensional equation for ERNN architecture and the integration of PSO in ERNN learning to produce the PSO-ERNN model. The PSO is constructed to train the NN and ERNN models to achieve a fast convergence rate and avoid local minima problems. The PSO-ERNN model is validated by comparing it against the standard PSO-NN metaheuristic model and similar models from the literature. The PSO-NN and PSO-ERNN models are tested and evaluated using ten benchmark classification problems of breast cancer, heart, hepatitis, liver, wine, iris, lung cancer, yeast, Pima Indians diabetes, and ionosphere datasets. In the training phase, the results show that the PSO-ERNN model performs better than the PSO-NN model when the training set has a bigger size of samples. In the testing phase, the PSO-ERNN model outperforms the PSO-NN model for all the tested datasets except the lung cancer and yeast datasets, in which the accuracy percentage slightly decreases. In the validation phase, the PSO-ERNN model shows better performance quality in terms of accuracy percentage in six of the tested datasets. The average percentage of the training, testing, and validation accumulation show that the PSO-NN performs better than the PSO-ERNN in the lung cancer (87.27, 83.32), and heart (73.56, 70.64) datasets. On the other hand, the PSO-ERNN performs better than the PSO-NN in the iris (88.18, 86.74), hepatitis (88.60, 87.93), wine (89.16, 86.08), liver (73.56, 70.64), ionosphere (83.98, 78.94), and breast cancer (94.84, 91.17). PSO-NN and PSO-ERNN produce the same average results in the Pima Indians diabetes (84.00, 84.00) and yeast (91.31, 91.30) dataset. These results show clearly that the PSO-ERNN generally outperforms the PSO-NN when considering the overall results of the ten datasets. Nevertheless, the combinations of the PSO-NN and PSO-ERNN are proven to represent consistent and robust classification methods. The computational efficiencies of the training processes in both the PSO-NN and PSO-ERNN models are highly improved when coupled with the PSO. [ABSTRACT FROM AUTHOR]

This paper presents a new method for initializing weights in a Feedforward Neural Network (FNN) with a single hidden layer combined with a constructive approach to define the number of hidden units associated with the best classification performance. The strategy consists of defining an initial number of hidden units according to the classification problem, the linearization of the whole network around an equilibrium point and the determination of the initial weights and bias through the maximum approximation of the linearized model to the Optimal Linear Classifier (OLC) whose solution can be obtained analytically. The constructive algorithm comprises a gradual increase in the number of hidden units in such a way that at each training only the weights and bias associated with the new hidden units are initiated randomly while the weights and bias obtained from previous training are used as initial guesses. Additionally, the constructive algorithm seeks to ensure that the loss function of the trained networks decreases with the successive additions of hidden units. The proposed approach (Weight Initialization based on the Linearization of the Whole Neural Network combined with a new Constructive Algorithm, WILWNN-CA) is applied to synthetic and real datasets widely used as benchmark for multi-class classification problems. The comparison with conventional random weight initialization and other approaches involving different network topologies (and initialization strategies) shows that the proposed method is efficient and capable of providing success rates (correct classification rates) higher or similar to those achieved with existing methods. • Novel approach for initializing weights in a Single Feedforward Neural Network. • Novel constructive algorithm to define the number of hidden units. • The method performed better or similarly to other consolidated methods. • Constructive approach ensures reduction of the loss function. • The weight Initialization is based on the linearization of the NN as a whole. [ABSTRACT FROM AUTHOR]