Search

Covid19 is a horrible disease, which upset our life everywhere. The main complexity of this disease lies in its rapid evolution and through people’s contact and gaps in our understanding. Moreover, it represents critical cases when the immune system has not presented any symptoms. Hence, the design of an effective classifier is necessary. This paper aims to hybrid the multi-criteria Analytic Hierarchy Pprocess (AHP) tool and the process of the convolutional neural network (CNN), for making the classification of a category of patients. Our novel method is divided into two main phases: the first one focuses on the generation of the priorities of the essential criteria using the AHP model, while the second phase aims to classify the patients using the neural network classifier. In the present study, we considered three important criteria: fever, patient, localiztion, and the age of the patient. From the obtained results, the proposed model has proved its efficiency even if we consider different cases.

Presently, the usage of medicinal images has drastically increased and it provides extensive details related to the patient’s health status. It shows the applicability of diagnosing the disease and stored in a memory for examination purposes. For the retrieval of medical images in a real world environment, significant need is present in the designing of an effective medical image indexing and retrieval technique. This paper offers an efficient medical image retrieval (MIR) model through feature extraction based classification model. Here, Directional local ternary quantized extrema patterns (DLTerQEPs) and autocorrelogram (AC) based feature extraction process takes place to extract texture and color features. Next, neural network (NN) based classification process takes place. The investigation of the simulation results takes place to showcase the betterment of the presented mode. During experimentation process, it is noticed that the presented model is superior to compared methods.

In this chapter we will consider another application case of Deep learning: classification of brain images for detection of Alzheimer’s disease. In this particular application of medical imaging domain, Deep NNs have become the mandatory tool. In this chapter we give some highlights on how the usual steps in design of a Deep Neural Network classifier are implemented in the case when domain knowledge has to be considered. But more than that: faithful to our principle of showing new aspects of Deep Learning even in application cases, we will show how information fusion with siamese CNNs helps in increasing of performances of these classifiers.

Alcoholism involves psychological and biological components where multiple risk factors come into play. Assessment of the psychiatric emergency is a challenging issue for clinicians working with alcohol-dependent patients. Identifying alcoholics from healthy controls from their EEG signals can be effective in this scenario. In this research, we have applied two instance-based classifiers and three neural network classifier to classify Electroencephalogram data of alcoholics and normal person. For data preprocessing, we have applied discrete wavelet transform, Principal component analysis and Independent component analysis. After successful implementation of the classifiers, an accuracy of 95% is received with Bidirectional Long Short-Term Memory. Finally, comparing the performance of the two categories of algorithms, we have found that neural networks have higher potentiality against instance-based classifiers in the classification of EEG signals of alcoholics.

Adverse Drug-Drug Interactions (DDIs) are a very important risk factor in the medical process, which may lead to readmission or death. Although a part of DDIs can be obtained through in vitro or in vivo experiments in the drug development stage, a large number of new DDIs still appear after the market, more and more researchers begin to pay attention to the research related to drug molecules, such as drug discovery, drug target prediction, DDIs prediction, etc. In recent years, many computational methods for predicting DDIs have been proposed. However, most of them only used labeled data and neglect a lot of information hidden in unlabeled data. Moreover, they always focus on binary prediction instead of multiclass prediction, although the exact DDI type is very helpful for our reasonable choice of medication. In this paper, a Semi-Surpervised Variational Autoencoders (SPRAT) method for predicting DDIs is proposed, which is composed of a neural network classifier and a Variational autoencoders (VAE). Classifier is the core components, VAE plays a role of calibration. In the end, the predicted label is a multi-hot vector which indicates specific DDI types between drug pairs. Finally, the experiments on real world dataset demonstrate the effectiveness of the proposed method in this paper.

Nearest neighbor partitioning (NNP) method has been proved to be an effective method to enhance the quality of neural network classifiers. However, there are many cluster shapes in NNP, which results in a large number of local optimal solutions in the searching space by the traditional particle swarm optimization (PSO) algorithm. Therefore, the multi-layer particle swarm optimization (MLPSO) is introduced to increase the diversity of searching groups through increasing the number of layers, thereby improving the performance when facing with large scale problems. In this study, we adopt the combination of multi-layer particle swarm optimization and nearest neighbor partitioning to solve the local optimal problem caused by multi-cluster shapes in the optimization of NNP. Experimental results show that this method improves the performance of classifier.

Floating centroid method (FCM) offers an efficient way to solve a fixed-centroid problem for the neural network classifiers. However, evolutionary computation as its optimization method restrains the FCM to achieve satisfactory performance for different neural network structures, because of the high computational complexity and inefficiency. Traditional gradient-based methods have been extensively adopted to optimize the neural network classifiers. In this study, a gradient-based floating centroid (GDFC) method is introduced to address the fixed centroid problem for the neural network classifiers optimized by gradient-based methods. Furthermore, a new loss function for optimizing GDFC is introduced. The experimental results display that GDFC obtains promising classification performance than the comparison methods on the benchmark datasets.

Within this paper, we evaluate two main machine learning techniques for credit scoring. The first algorithm consists of a cascade with two steps: (a) C4.5 decision tree; (b) AdaBoost for binary classification (credit accepted or rejected). The second technique corresponds to choosing a neural network classifier implemented by Multilayer Perceptron (MLP). For evaluation of the proposed models, we have used the German credit dataset and the Australian credit dataset. For the German dataset, MLP leads to the best result corresponding to an accuracy of 81.0%, versus C4.5 enhanced with AdaBoost that leads to an accuracy of 78.67%. For the Australian credit dataset, we found that MLP is also the best classifier with an accuracy of 90.85%, versus C4.5 followed by AdaBoost obtaining an accuracy of 89.00%. At the same time, one can remark that C4.5 enhanced by AdaBoost has led to a better performance than a simple C4.5 .

Machine learning models, especially neural network (NN) classifiers, are widely used in many applications including natural language processing, computer vision and cybersecurity. They provide high accuracy under the assumption of attack-free scenarios. However, this assumption has been defied by the introduction of adversarial examples – carefully perturbed samples of input that are usually misclassified. Many researchers have tried to develop a defense against adversarial examples; however, we are still far from achieving that goal. In this paper, we design a Generative Adversarial Net (GAN) based adversarial training defense, dubbed GanDef, which utilizes a competition game to regulate the feature selection during the training. We analytically show that GanDef can train a classifier so it can defend against adversarial examples. Through extensive evaluation on different white-box adversarial examples, the classifier trained by GanDef shows the same level of test accuracy as those trained by state-of-the-art adversarial training defenses. More importantly, GanDef-Comb, a variant of GanDef, could utilize the discriminator to achieve a dynamic trade-off between correctly classifying original and adversarial examples. As a result, it achieves the highest overall test accuracy when the ratio of adversarial examples exceeds 41.7%.

Layer-wise relevance propagation (LRP) has shown potential for explaining neural network classifier decisions. In this paper, we investigate how LRP is to be applied to deep neural network which makes use of batch normalization (BatchNorm), and show that despite the functional simplicity of BatchNorm, several intuitive choices of published LRP rules perform poorly for a number of frequently used state of the art networks. Also, we show that by using the \(\varepsilon \)-rule for BatchNorm layers we are able to detect training artifacts for MobileNet and layer design artifacts for ResNet. The causes for such failures are analyzed deeply and thoroughly. We observe that some assumptions on the LRP decomposition rules are broken given specific networks, and propose a novel LRP rule tailored for BatchNorm layers. Our quantitatively evaluated results show advantage of our novel LRP rule for BatchNorm layers and its wide applicability to common deep neural network architectures. As an aside, we demonstrate that one observation made by LRP analysis serves to modify a ResNet for faster initial training convergence.

In this paper, MSA speech rhythm metrics were used to classify two regional accent (northern vs. southern regions) using an MLP - neural network classifier. Seven rhythm metrics vectors were computed from a speech dataset taken from ALGerian Arabic Speech Database (ALGASD) using both Interval Measures (IM) and Control/Compensation Index (CCI) algorithms. The classifier was trained and tested using different input vectors of speech rhythm measurements. The best accuracy of the NN-classifier was achieved when a combination of all metrics was used (88.6%).

Aim of the present paper is the analysis of the strain along the beam that is equipped with Glass Fibers Reinforced Polymers (GFRP) with an embedded set of optical Fiber Bragg Grating sensors (FBG), in the context of a project to equip with these new structural elements an Italian train bridge.

Intent detection is one of the main tasks of a dialogue system. In this paper we present our intent detection system that is based on FastText word embeddings and neural network classifier. We find a significant improvement in the FastText sentence vectorization. The results show that our intent detection system provides state-of-the-art results on three English datasets outperforming many popular services.

This paper evaluates the capability of the hybrid parameter reduction approach in handling private financial initiative (PFI) unitary charges data to increase the classification performance. The objective of this study is to analyse the performance of the proposed hybrid parameter reduction approach in assisting the neural network classifier to classify complex data sets that might contain uncertain and inconsistent problems. The proposed hybrid parameter reduction approach consists of several methods that will be executed during the data analysis process. Slicing technique and dominance-based rough set approach (DRSA) are the two techniques that play important roles in the proposed parameter reduction process. In order, to analyse the performance of the proposed work, the PFI data that covers all regions in Malaysia is applied in the experimental works. Besides, several standard data sets have also been used to validate the obtained results. The results reveal that the hybrid approach has successfully assisted the classifier in the classification process.

The appearance of spot welding reflects the quality of welding to a large extent. In this study, we developed a vision inspection system, which recognizes the defects of weld in electronic components based on neural network. First, the images of weld are acquired by color camera. Then, we extracted 15 features from the welding images that had been corrected and enhanced. Finally, we used 1800 training samples to train the neural network. And then we got a accuracy of 95.82% under 407 testing samples by the neural network classifier, which had 15 input nodes, 4 hidden nodes and 2 output nodes.

Speaker discrimination consists in checking whether two (or more) speech segments belong to the same speaker or not. In this framework, we propose a new approach developed for the task of speaker discrimination, this approach results from the fusion between a neural network classifier (NN) and a statistical classifier, this fusion is obtained once by combining the scores of the simple classifiers weighted by some confidence coefficients and another time, by using the scores of the statistical classifier as an additional input of the Multi-Layer Perceptron (MLP), in order to optimize the NN training (Hybrid model).

This paper investigates how an intelligent agent could be designed to both predict whether it is bonding with its user, and convey appropriate facial expression and body language responses to foster bonding. Video and Kinect recordings are collected from a series of naturalistic conversations, and a reliable measure of bonding is adapted and verified. A qualitative and quantitative analysis is conducted to determine the non-verbal cues that characterize both high and low bonding conversations. We then train a deep neural network classifier using one minute segments of facial expression and body language data, and show that it is able to accurately predict bonding in novel conversations.

In Albanian language there are several phonemes that are similar in pronunciation like /q/ - /c/, /rr/ - /r/, /th/ - /dh/ and /gj/ - /xh/. These phonemes are difficult to distinguish by human ear even for native speaking Albanians from different regions. The task becomes more challenging for automated speech systems, recognizing and classifying Albanian words and language due to the similar sounding phonemes. This paper proposes to use Mel Frequency Cepstral Coefficients (MFCC) based features to distinguish these phonemes correctly. The three layers back propagation neural network is used for classification. The experiments are performed on speech signals that are collected from different male and female native speakers. The speaker independent tests are performed for analyzing the performance of the classification. The obtained results show that the serial MFCC features can be used to classify the very similar speech phonemes with higher accuracy.

International audience; Accurate quantification of white matter hyperintensities (WMH) from MRI is a valuable tool for studies on ageing and neurodegeneration. Reliable automatic extraction of WMH biomarkers is challenging, primarily due to their heterogeneous spatial occurrence, their small size and their diffuse nature. In this paper, we present an automatic and accurate method to segment these le-sions that is based on the use of neural networks and an overcomplete strategy. The proposed method was compared to other related methods showing competitive and reliable results in two different neurodegenerative datasets.

In an agricultural country like India, majority of population depend on plant produce for their survival. Plants occupy a large portion of our ecosystem. In order to derive different benefits from plants in an optimum manner, one needs to be aware of the properties being possessed by plants. For that purpose, one needs to have proper source carrying significant information about plants and an expert so as to respond to ones queries. However, both these are not available in adequate which drives the need to create automation in the process of recognition of leaves for plant classification. Thus, a novel algorithm has been developed which helps in recognizing different varieties of leaves without human interference. The system uses real time images of leaves and extracts physiological as well as morphological features of the leaves, which are then fed as input to a classifier. The same has been implemented on a Back propagation based neural network classifier and a comparative study has been made. The study shows that the recognition rates of the proposed method are more accurate than that of BPNN and the proposed algorithm is found to be an efficient one.

In this work, principles of operation, advantages and disadvantages are presented for different detector technologies. An idea of a new detection and classification method for a single magnetic sensor based system is also discussed. It is important that the detection algorithm and the neural network classifier needs to be easily implementable in a microcontroller based system.

The act of controlling pressure through pencil and brush appears effortless, but to mimic this natural ability in the realm of electronic medium using tablet pen device is difficult. Previous pressure based interaction work have explored various signal processing techniques to improve the accuracy in pressure control, but a one-for-all signal processing solutions tend not to work for different curve types. We propose instead a framework which applies signal processing techniques tuned to individual curve type. A neural network classifier is used as a curve classifier. Based on the classification, a custom combination of signal processing techniques is then applied. Results obtained point to the feasibility and advantage of the approach. The results are generally applicable to the design of pressure based interaction technique and possibly unlock the potential of pressure based system for richer interactions.

This paper is focused on evaluating the capability of MLP and RBF neural network classifier algorithms for performing handwritten character recognition task. Projection profile features for the character images are extracted and merged with the binarization features obtained after preprocessing every character image. The fused features thus obtained are used to train both the classifiers i.e. MLP and RBF Neural Networks. Simulation studies are examined extensively and the proposed fused features are found to deliver better recognition accuracy when used with RBF Network as a classifier.

In this paper a proposal for solving the problem of diagnostics of cutting errors in a rotary cutoff in a corrugated board machine processing line is presented. There are many different reasons for errors, and their identification requires a sound knowledge and experience of the service staff. The authors, using their many years’ experience and a huge database, have found that many sources of errors can be characterized using a probability density function (pdf). They proposed a diagnostics method based on classification of sources of disturbances using the analysis of pdf determined with a kernel density estimator. Multilayer feedforward neural network is proposed as a classifier. Classification procedure is discussed, together with research results based on data from real industrial processes.