Vector Size Research Articles

A novel framework of MOPSO-GDM in recognition of Alzheimer's EEG-based functional network.

Most patients with Alzheimer's disease (AD) have an insidious onset and frequently atypical clinical symptoms, which are considered a normal consequence of aging, making it difficult to diagnose AD medically. But then again, accurate diagnosis is critical to prevent degeneration and provide early treatment for AD patients. This study aims to establish a novel EEG-based classification framework with deep learning methods for AD recognition. First, considering the network interactions in different frequency bands (δ, θ, α, β, and γ), multiplex networks are reconstructed by the phase synchronization index (PSI) method, and fourteen topology features are extracted subsequently, forming a high-dimensional feature vector. However, in feature combination, not all features can provide effective information for recognition. Moreover, combining features by manual selection is time-consuming and laborious. Thus, a feature selection optimization algorithm called MOPSO-GDM was proposed by combining multi-objective particle swarm optimization (MOPSO) algorithm with Gaussian differential mutation (GDM) algorithm. In addition to considering the classification error rates of support vector machine, naive bayes, and discriminant analysis classifiers, our algorithm also considers distance measure as an optimization objective. Finally, this method proposed achieves an excellent classification error rate of 0.0531 (5.31%) with the feature vector size of 8, by a ten-fold cross-validation strategy. These findings show that our framework can adaptively combine the best brain network features to explore network synchronization, functional interactions, and characterize brain functional abnormalities, which can improve the recognition efficiency of diseases. While improving the classification accuracy of application algorithms, we aim to expand our understanding of the brain function of patients with neurological disorders through the analysis of brain networks.

Stellar clustering and the kinematics of stars around Collinder 121 using Gaia DR3

ABSTRACT We study the region around Collinder 121 (Cr 121) using newly available 6D data from the Gaia DR3 catalogue. Situated in the third quadrant, near the Galactic plane, Collinder 121 lies in the region of Canis Major centred around (l, b) ≈(236°, −10°). Previous studies have suggested that the stellar associations in this region comprise an OB association (CMa OB2) lying at about 740 pc with a more distant open cluster (Cr 121) at approximately 1170 pc. Despite these studies, the precise nature of Collinder 121 remains uncertain. This study investigates the region bounded by the box (l, b) = (225°–245°, 0.00° to −20.00°) to a depth of 700 pc from 500 to 1200 pc which fully encompasses the region discussed in the literature. Using Gaia DR3 data, we do not find associations at the distances given in the literature. Instead, using the hdbscan machine learning algorithm, we find a major association of OB stars centred around 803 pc. Within this association we find four smaller subgroups that may be indicative of a larger association and which are located at a mean distance of 827 pc. Proper motion studies find coherence between these four subgroups and show a distinctive east-to-west increase in the size of the velocity vectors which supports contemporary studies that show similar trends in OB populations in Cygnus and within the Carina spiral Arm. Therefore, we hypothesize that Cr 121 and CMa OB2 are the same cluster, consistent with the 1977 study by Hoogerwerf.

Automatic depression severity assessment with deep learning using parameter-efficient tuning.

To assist mental health care providers with the assessment of depression, research to develop a standardized, accessible, and non-invasive technique has garnered considerable attention. Our study focuses on the application of deep learning models for automatic assessment of depression severity based on clinical interview transcriptions. Despite the recent success of deep learning, the lack of large-scale high-quality datasets is a major performance bottleneck for many mental health applications. A novel approach is proposed to address the data scarcity problem for depression assessment. It leverages both pretrained large language models and parameter-efficient tuning techniques. The approach is built upon adapting a small set of tunable parameters, known as prefix vectors, to guide a pretrained model towards predicting the Patient Health Questionnaire (PHQ)-8 score of a person. Experiments were conducted on the Distress Analysis Interview Corpus - Wizard of Oz (DAIC-WOZ) benchmark dataset with 189 subjects, partitioned into training, development, and test sets. Model learning was done on the training set. Prediction performance mean and standard deviation of each model, with five randomly-initialized runs, were reported on the development set. Finally, optimized models were evaluated on the test set. The proposed model with prefix vectors outperformed all previously published methods, including models which utilized multiple types of data modalities, and achieved the best reported performance on the test set of DAIC-WOZ with a root mean square error of 4.67 and a mean absolute error of 3.80 on the PHQ-8 scale. Compared to conventionally fine-tuned baseline models, prefix-enhanced models were less prone to overfitting by using far fewer training parameters (<6% relatively). While transfer learning through pretrained large language models can provide a good starting point for downstream learning, prefix vectors can further adapt the pretrained models effectively to the depression assessment task by only adjusting a small number of parameters. The improvement is in part due to the fine-grain flexibility of prefix vector size in adjusting the model's learning capacity. Our results provide evidence that prefix-tuning can be a useful approach in developing tools for automatic depression assessment.

Performance Assessment of Joint Optical-Digital Nonlinearity Mitigation Schemes in Long-Haul Systems

The combined approach of optical and digital nonlinearity mitigation techniques have been shown to have an edge over the individual schemes, tackling their drawbacks and improving compensation capabilities. We demonstrated fiber-nonlinearity mitigation of 32 GBd single-polarization 16QAM signals transmitted over an 800-km dispersion-managed link using optical phase conjugation (OPC) and digital-domain neural networks (NN). Our implemented NN comprised a recurrent neural network (RNN), and a reservoir computing network (RCN). To further compensate for the penalty introduced by the design of OPC, we employed NN schemes in the digital signal processing and applied it to the signal after transmission in the OPC-assisted link. We present the results for optimizing the NN models for important hyper-parameters like the number of hidden layer neurons and the input vector size. The proposed joint approach achieves Q <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> -factor improvements up to 1.8 dB while surpassing the improvements of the individual schemes. Our experiments indicate that the joint approach has the potential to reduce the overall complexity of NN architectures in terms of the size of the hidden layer and input vector.

Categorization of Arabic posts using Artificial Neural Network and hash features

Sentiment analysis is an important study topic with diverse application domains including social network monitoring and automatic analysis of the body of natural language communication. Existing research on sentiment analysis has already utilised substantial domain knowledge available online comprising users’ opinion in various areas such as business, education, and social media. There is however limited literature available on Arabic language sentiment analysis. Furthermore, datasets used in majority of these studies have poor classification. In the present study, we utilised a primary dataset comprising 2122 sentences and 15,331 words compiled from 206 publicly available online posts to perform sentiment classification by using advanced machine learning technique based on Artificial Neural Networks. Unlike lexicon-based techniques that suffer from low accuracy due to their computational nature and parameter configuration, Artificial Neural Networks were used to classify people opinion posts into three categories including conservative, reform and revolution, accompanied by multiple hasher vector size to benchmark the performance of the proposed model. Extensive simulation results indicated an accuracy of 93.33%, 100%, and 100% for the classification of conservation, reform, and revolutionary classes, respectively.

Texture Detection With Feature Extraction on Embedded FPGA

A feature extraction algorithm for texture detection oriented to its implementation on embedded electronics based on a Field-Programmable Gate Array (FPGA) is proposed in this paper. Local pre-processing with smart tactile sensors can help to improve dexterity in artificial hands. Simplicity is the goal in order to achieve a hardware-friendly strategy that can be replicated and integrated with another circuitry. This is interesting, considering that tactile sensors are arrays and FPGAs are capable of parallel execution. The proposal was tested with a custom smart tactile sensor mounted on a Cartesian robot to explore different textures. A comparison with a common feature extraction approach based on the fast Fourier Transform (FFT) computation was also made. In addition, the whole procedure is implemented on a System on Chip (SoC) with the feature extraction on the embedded FPGA and a <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">k-means</i> classifier on an ARM core. The proposed algorithm obtains the spatial frequency components of the tactile signal but not their power. Therefore, some information is lost with respect to that provided by the FFT. Nevertheless, an 89.17% accuracy of the proposed algorithm is obtained versus 91.4% with the FFT when twelve different textures are considered, including complex and fabric textures. There is a noticeable saving in power and hardware resources. In addition, since the size of the feature vector is much smaller, data traffic and memory usage is much lower, and the classifier can be simpler.

A novel machine learning-based feature extraction method for classifying intracranial hemorrhage computed tomography images

One of the most serious forms of brain stroke is intracranial hemorrhage (ICH). When an artery bursts, the brain and the tissue around the artery start bleeding. This study proposes a joint feature selection strategy to classify computed tomography (CT) images of intracranial hemorrhage. The joint feature set is composed of transform and texture features. Joint features are constructed from a combination of grey level co-occurrence matrix (GLCM) features, discrete wavelet features (DWT), and discrete cosine features (DCT). Brain hemorrhage CT image classification uses ensemble-based machine learning (ML) techniques. On the training dataset, a Synthetic Minority Over-Sampling Technique (SMOTE) is applied to treat the problem of oversampling by adding fresh data. Additionally, the sequential forward feature selection technique is used to obtain feature subsets. The classification accuracy is further examined for varied feature vector sizes. Confusion matrix, precision, and recall in categorization are employed as performance evaluation measurements. The ML-based ensemble classifiers can produce highly accurate results with the aid of the proposed novel feature extraction mechanism. When taking into consideration a crucial feature set consisting of six features, it can be seen that Random Forest obtained the greatest accuracy, which is 87.22%.

Leukocytes Classification for Leukemia Detection Using Quantum Inspired Deep Feature Selection.

Leukocytes, also referred to as white blood cells (WBCs), are a crucial component of the human immune system. Abnormal proliferation of leukocytes in the bone marrow leads to leukemia, a fatal blood cancer. Classification of various subtypes of WBCs is an important step in the diagnosis of leukemia. The method of automated classification of WBCs using deep convolutional neural networks is promising to achieve a significant level of accuracy, but suffers from high computational costs due to very large feature sets. Dimensionality reduction through intelligent feature selection is essential to improve the model performance with reduced computational complexity. This work proposed an improved pipeline for subtype classification of WBCs that relies on transfer learning for feature extraction using deep neural networks, followed by a wrapper feature selection approach based on a customized quantum-inspired evolutionary algorithm (QIEA). This algorithm, inspired by the principles of quantum physics, outperforms classical evolutionary algorithms in the exploration of search space. The reduced feature vector obtained from QIEA was then classified with multiple baseline classifiers. In order to validate the proposed methodology, a public dataset of 5000 images of five subtypes of WBCs was used. The proposed system achieves a classification accuracy of about 99% with a reduction of 90% in the size of the feature vector. The proposed feature selection method also shows a better convergence performance as compared to the classical genetic algorithm and a comparable performance to several existing works.

Metaheuristic Optimization-Based Feature Selection for Imagery and Arithmetic Tasks: An fNIRS Study.

In recent decades, the brain-computer interface (BCI) has emerged as a leading area of research. The feature selection is vital to reduce the dataset's dimensionality, increase the computing effectiveness, and enhance the BCI's performance. Using activity-related features leads to a high classification rate among the desired tasks. This study presents a wrapper-based metaheuristic feature selection framework for BCI applications using functional near-infrared spectroscopy (fNIRS). Here, the temporal statistical features (i.e., the mean, slope, maximum, skewness, and kurtosis) were computed from all the available channels to form a training vector. Seven metaheuristic optimization algorithms were tested for their classification performance using a k-nearest neighbor-based cost function: particle swarm optimization, cuckoo search optimization, the firefly algorithm, the bat algorithm, flower pollination optimization, whale optimization, and grey wolf optimization (GWO). The presented approach was validated based on an available online dataset of motor imagery (MI) and mental arithmetic (MA) tasks from 29 healthy subjects. The results showed that the classification accuracy was significantly improved by utilizing the features selected from the metaheuristic optimization algorithms relative to those obtained from the full set of features. All of the abovementioned metaheuristic algorithms improved the classification accuracy and reduced the feature vector size. The GWO yielded the highest average classification rates (p < 0.01) of 94.83 ± 5.5%, 92.57 ± 6.9%, and 85.66 ± 7.3% for the MA, MI, and four-class (left- and right-hand MI, MA, and baseline) tasks, respectively. The presented framework may be helpful in the training phase for selecting the appropriate features for robust fNIRS-based BCI applications.

Face presentation attack detection performances of facial regions with multi-block LBP features.

Biometric recognition systems are frequently used in daily life although they are vulnerable to attacks. Today, especially the increasing use of face authentication systems has made these systems the target of face presentation attacks (FPA). This has increased the need for sensitive systems detecting the FPAs. Recently surgical masks, frequently used due to the pandemic, directly affect the performance of face recognition systems. Researchers design face recognition systems only from the eye region. This motivated us to evaluate the FPA detection performance of the eye region. Based on this, in cases where the whole face is not visible, the FPA detection performance of other parts of the face has also been examined. Therefore, in this study, FPA detection performances of facial regions of wide face, cropped face, eyes, nose, and mouth was investigated. For this purpose, the facial regions were determined and normalized, and texture features were extracted using powerful texture descriptor local binary patterns (LBP) due to its easy computability and low processing complexity. Multi-block LBP features are used to obtain more detailed texture information. Generally uniform LBP patterns are used for feature extraction in the literature. In this study, the FPA detection performances of both uniform LBP patterns and all LBP patterns were investigated. The size of feature vector is reduced by principal component analysis, and real/fake classification is performed with support vector machines. Experimental results on NUAA, CASIA, REPLAY-ATTACK and OULU-NPU datasets show that the use of all patterns increased the performance of FPA detection.

Design of a modern fast Fourier transform and cache effective bit-reversal algorithm

ABSTRACT This article deals with efficient vectorization of the fast Fourier transform algorithm while focusing on Cooley–Tukey versions with power-of-two radixes. Aside from examples of optimizations for 256 and 512-bit vectors, this work also discusses relations between individual radix-based versions, vectorization and OpenMP threading. Ideas are progressing into a timeless design of the FFT algorithm, which can work with any vector size and radix version through conversion into radix-2 output permutation. Furthermore, the implementation of the Cache Optimized Bit-Reversal algorithm, which doubles the performance of its predecessor, is introduced.

Correction: Temporal and spatially controlled APP transgene expression using Cre-dependent alleles

There were errors in Disease Models & Mechanisms (2022) 15, dmm049330 (doi:10.1242/dmm.049330).The authors recently discovered that one of the Cre-dependent transgenic mouse lines used in the study is not correctly described within the article. The mouse line labelled as R26-APP in the paper was originally described as containing both the Swedish and Iberian mutations, but carries only the Iberian mutation. This was discovered by The Jackson Laboratory when they sequenced through the complete insert to confirm the allele before entering the line into their repository for distribution. The authors also discovered an unintended point mutation located just outside the Aβ domain, M647T by APP695 numbering (M722T by APP770 numbering). They propose that the model should be identified as R26-APPIbe rather than as R26-APP.A diagram of the allele that appears in Fig. 2A has been corrected:In addition, a section describing the cloning methods in the Materials and Methods has been updated. The corrected and original versions are shown below:‘Cloning of R26-LSL-hAPPIbe transgene’ section (corrected). Human APPIbe 695 amino acid cDNA was generated in two steps. The N-terminal 1592 nucleotides of huAPP695 were cut from pcDNA3.1-sAPPα (a gift from Hui Zheng, BCM) with HindIII and SacI and subcloned into the same sites of transfer plasmid L4440 (Addgene, #1654). The remaining fragments were amplified from a human cDNA library using a two-step PCR method. Both PCR products contained the Ibe mutation as well as an overlap region needed to serve as an extended template for the second PCR. The first C-terminal fragment was amplified using hAPP C-terminus forward primer 5′-AACGAAAACCACCGTGGAGCT-3′ and Ibe reverse primer 5′-AAGGTGATGACGAACACTGTCG-3′. The second C-terminal fragment was amplified by Ibe forward primer 5′-CGACAGTGTTCGTCATCACCTT-3′ and hAPP C-terminus reverse primer 5′-GCGCGGATATCCTAGTTCTGCATCTGCTCAAAG-3′. The complete hAPP695Ibe C-terminus was then amplified from a mixture of the two shorter products using hAPP CT forward primer and hAPP CT reverse primer. The purified huAPPIbe C-terminal PCR fragment was digested with EcoRV and SacI and ligated into the EcoRV/SacI sites of L4440-APP(1-1592), resulting in L4440-APPIbe. The complete huAPPIbe sequence was transferred into an intermediary expression plasmid, which was digested with MluI and FseI and cloned into an Ai3-derived ROSA26 targeting vector (Madisen et al., 2010) to generate the final transgene construct pJJ-hAPPIbe. The modified Ai3 targeting vector retained the original Ai3 Cre-dependent CAG-loxP-stop-loxP sequence but removed the PGK-DTA negative selection cassette and shortened the 3′ targeting arm to reduce the overall vector size. All restriction enzymes were purchased from New England Biolabs (Ipswich, MA, USA).‘Cloning of R26-LSL-hAPPSwe/Ibe transgene’ section (original). Human APPSwe/Ibe 695 amino acid cDNA was amplified from the pBR322-APPSwe/Ind plasmid used to generate the Tg CRND8 mouse model (Chishti et al., 2001) (a gift from David Westaway, University of Alberta, Canada) with primers that converted the original Indiana mutation V717F into the Iberian mutation I716F (point mutation numbering is for the standard human APP 770 amino acid isoform). The N-terminus of APPSwe/Ibe 695 was amplified using the forward primer 5′-GGCCGGCCGCCACCATGCTGCCCGGTTTGGCACT-3′ and reverse primer 5′-AAGGTGATGACGAACACTGTCG-3′; the C-terminus was amplified using the forward primer 5′-CGACAGTGTTCGTCATCACCTT-3′ and reverse primer 5′-GCGCGACGCGTCTAGTTCTGCATCTGCTCAAAGAA-3′. The resulting hAPPSwe/Ibe fragments were combined by amplification using the forward primer 5′-GGCCGGCCGCCACCATGCTGCCCGGTTTGGCACT-3′ and reverse primer 5′-GCGCGACGCGTCTAGTTCTGCATCTGCTCAAAGAA-3′. The complete hAPPSwe/Ibe 695 cassette was digested with MluI and FseI, and cloned into an Ai3-derived ROSA26 targeting vector (Madisen et al., 2010) that had been cut with the same enzymes to generate the final transgene construct pJJ-hAPPSwe/Ibe. The modified Ai3 targeting vector retained the original Cre-dependent CAGloxP-stop-loxP sequence but removed the PGK-DTA negative selection cassette and shortened the 3′ targeting arm to reduce the overall vector size. All restriction enzymes were purchased from New England Biolabs (Ipswich, MA, USA).Throughout the paper, where the text refers to R26-APP, R26-APPSwe/Ibe or ROSA26-CAG-LSL-APPSwe/Ibe, corrections have been made to R26-APPIbe or ROSA26-CAG-LSL-APPIbe. The model will still be distributed by The Jackson Laboratory as stock #037319, but using the line name R26-APPIbe. The study results and conclusions remain unaffected.Both the online full-text and PDF versions have been updated. The authors apologise to readers for these errors and any inconvenience they may have caused.

Cover Picture: Astron. Nachr. 03/2023

Vector plots of the Geometric distortion (GD) solution for the James Webb Telescope NIRCam long wavelength channel detectors #5 of both module A (left) and module B (right), showing the size of the residual vectors, magnified by a factor of 20. The color map represents the pixels' area variation across the detectors due to the GD correction. Details are presented in the article by M. Griggio et al. (this issue of AN, e230006) where the authors describe their correction procedure and its validation. Along with their geometric-distortion solution, they make available their Python software to convert the raw-pixels coordinates of each detector into distortion-free positions, and to put all ten detectors of NIRCam into a common reference system.

An efficiency-driven, correlation-based feature elimination strategy for small datasets

With big datasets and highly efficient algorithms becoming increasingly available for many problem sets, rapid advancements and recent breakthroughs achieved in the field of machine learning encourage more and more scientific fields to make use of such a computational data analysis. Still, for many research problems, the amount of data available for training a machine learning (ML) model is very limited. An important strategy to combat the problems arising from data sparsity is feature elimination—a method that aims at reducing the dimensionality of an input feature space. Most such strategies exclusively focus on analyzing pairwise correlations, or they eliminate features based on their relation to a selected output label or by optimizing performance measures of a certain ML model. However, those strategies do not necessarily remove redundant information from datasets and cannot be applied to certain situations, e.g., to unsupervised learning models. Neither of these limitations applies to the network-based, correlation-driven redundancy elimination (NETCORE) algorithm introduced here, where the size of a feature vector is reduced by considering both redundancy and elimination efficiency. The NETCORE algorithm is model-independent, does not require an output label, and is applicable to all kinds of correlation topographies within a dataset. Thus, this algorithm has the potential to be a highly beneficial preprocessing tool for various machine learning pipelines.

Augmenting the training database with the method of gradual similarity ratios in the face recognition systems

In face recognition systems, light direction, reflection, and emotional and physical changes on the face are some of the main factors that make recognition difficult. Deep metric learning algorithms called representative learning are frequently preferred in this field. However, in addition to the model's success in feature extraction, factors such as the distribution of samples in this database and appropriate classifier preferences also affect the overall performance of the face recognition system. This study it is aimed to create integrity in the database of a pre-trained deep neural network model by obtaining augmented data for classes with a limited number of samples. Thanks to this method called Graded Similarity Rates (GSR), augmented data that could disrupt class integrity has been removed from the database. This way, classes with limited examples are kept integrity, and classifier behavior is used more effectively. The model proposed in the experimental study reached 99.38% accuracy values compared to traditional data augmentation models. Experimental results have shown that the database has an acceptable level of success even at smaller vector sizes and is more organized.

Efficient Net-XGBoost: An Implementation for Facial Emotion Recognition Using Transfer Learning

Researchers are interested in Facial Emotion Recognition (FER) because it could be useful in many ways and has promising applications. The main task of FER is to identify and recognize the original facial expressions of users from digital inputs. Feature extraction and emotion recognition make up the majority of the traditional FER. Deep Neural Networks, specifically Convolutional Neural Network (CNN), are popular and highly used in FER due to their inherent image feature extraction process. This work presents a novel method dubbed as EfficientNet-XGBoost that is based on Transfer Learning (TL) technique. EfficientNet-XGBoost is basically a cascading of the EfficientNet and the XGBoost techniques along with certain enhancements by experimentation that reflects the novelty of the work. To ensure faster learning of the network and to overcome the vanishing gradient problem, our model incorporates fully connected layers of global average pooling, dropout and dense. EfficientNet is fine-tuned by replacing the upper dense layer(s) and cascading the XGBoost classifier making it suitable for FER. Feature map visualization is carried out that reveals the reduction in the size of feature vectors. The proposed method is well-validated on benchmark datasets such as CK+, KDEF, JAFFE, and FER2013. To overcome the issue of data imbalance, in some of the datasets namely CK+ and FER2013, we augmented data artificially through geometric transformation techniques. The proposed method is implemented individually on these datasets and corresponding results are recorded for performance analysis. The performance is computed with the help of several metrics like precision, recall and F1 measure. Comparative analysis with competent schemes are carried out on the same sample data sets separately. Irrespective of the nature of the datasets, the proposed scheme outperforms the rest with overall rates of accuracy being 100%, 98% and 98% for the first three datasets respectively. However, for the FER2013 datasets, efficiency is less promisingly observed in support of the proposed work.

White Blood Cells Classification Using Entropy-Controlled Deep Features Optimization.

White blood cells (WBCs) constitute an essential part of the human immune system. The correct identification of WBC subtypes is critical in the diagnosis of leukemia, a kind of blood cancer defined by the aberrant proliferation of malignant leukocytes in the bone marrow. The traditional approach of classifying WBCs, which involves the visual analysis of blood smear images, is labor-intensive and error-prone. Modern approaches based on deep convolutional neural networks provide significant results for this type of image categorization, but have high processing and implementation costs owing to very large feature sets. This paper presents an improved hybrid approach for efficient WBC subtype classification. First, optimum deep features are extracted from enhanced and segmented WBC images using transfer learning on pre-trained deep neural networks, i.e., DenseNet201 and Darknet53. The serially fused feature vector is then filtered using an entropy-controlled marine predator algorithm (ECMPA). This nature-inspired meta-heuristic optimization algorithm selects the most dominant features while discarding the weak ones. The reduced feature vector is classified with multiple baseline classifiers with various kernel settings. The proposed methodology is validated on a public dataset of 5000 synthetic images that correspond to five different subtypes of WBCs. The system achieves an overall average accuracy of 99.9% with more than 95% reduction in the size of the feature vector. The feature selection algorithm also demonstrates better convergence performance as compared to classical meta-heuristic algorithms. The proposed method also demonstrates a comparable performance with several existing works on WBC classification.

MBi-GRUMCONV: A novel Multi Bi-GRU and Multi CNN-Based deep learning model for social media sentiment analysis

Today, internet and social media is used by many people, both for communication and for expressing opinions about various topics in many domains of life. Various artificial intelligence technologies-based approaches on analysis of these opinions have emerged natural language processing in the name of different tasks. One of these tasks is Sentiment analysis, which is a popular method aiming the task of analyzing people’s opinions which provides a powerful tool in making decisions for people, companies, governments, and researchers. It is desired to investigate the effect of using multi-layered and different neural networks together on the performance of the model to be developed in the sentiment analysis task. In this study, a new, deep learning-based model was proposed for sentiment analysis on IMDB movie reviews dataset. This model performs sentiment classification on vectorized reviews using two methods of Word2Vec, namely, the Skip Gram and Continuous Bag of Words, in three different vector sizes (100, 200, 300), with the help of 6 Bidirectional Gated Recurrent Units and 2 Convolution layers (MBi-GRUMCONV). In the experiments conducted with the proposed model, the dataset was split into 80%-20% and 70%-30% training-test sets, and 10% of the training splits were used for validation purposes. Accuracy and F1 score criteria were used to evaluate the classification performance. The 95.34% accuracy of the proposed model has outperformed the studies in the literature. As a result of the experiments, it was found that Skip Gram has a better contribution to classification success.

A Random Projection Approach to Hypothesis Tests in High-Dimensional Single-Index Models

In this article, we consider the problem of hypothesis testing in high-dimensional single-index models. First, we study the feasibility of applying the classical F-test to a single-index model when the dimension of covariate vector and sample size are of the same order, and derive its asymptotic null distribution and asymptotic local power function. For the ultrahigh-dimensional single-index model, we construct F-statistics based on lower-dimensional random projections of the data, and establish the asymptotic null distribution and the asymptotic local power function of the proposed test statistics for the hypothesis testing of global and partial parameters. The new proposed test possesses the advantages of having a simple structure as well as being easy to compute. We compare the proposed test with other high-dimensional tests and provide sufficient conditions under which the proposed tests are more efficient. We conduct simulation studies to evaluate the finite-sample performances of the proposed tests and demonstrate that it has higher power than some existing methods in the models we consider. The application of real high-dimensional gene expression data is also provided to illustrate the effectiveness of the method. Supplementary materials for this article are available online.

D3SENet: A hybrid deep feature extraction network for Covid-19 classification using chest X-ray images

Covid-19 is one of the biggest global epidemics seen in the world in recent years. Because of this, people's daily lifestyles, the economic conditions of countries and individuals, and most importantly, their health status has been adversely affected all over the world. Millions of people around the world have died from this disease. For this reason, rapid and accurate detection of the disease is of great importance in terms of treatment and precautions. In addition, it is especially important to correctly distinguish between Covid-19 and non-Covid-19 pneumonia diseases for correct diagnosis and treatment. These two diseases cause similar symptoms, and the symptoms and the effects of the disease on the body should be carefully examined for their differentiation. Chest X-ray images, chest computerized tomography, and swab tests are commonly used to detect patients infected with COVID-19. This disease affects the lungs the most in the body and causes fatal side effects such as shortness of breath. Therefore, medical images taken from the chest play an important role in the diagnosis of the disease. The fact that X-rays are faster and cheaper than computerized tomography has led to an increase in studies on the detection of disease with X-rays.In recent years, the impressive results of deep learning in the field of computer vision have attracted researchers to this field when working with image data. This study aims to detect these diseases on chest X-ray images collected from Covid-19 patients, pneumonia patients, and healthy individuals. We proposed a hybrid feature extraction network namely D3SENET which consists of DarkNet53, DarkNet19, DenseNet201, SqueezeNet, and EfficientNetb0. After a balanced data set was prepared, feature vectors were obtained from images using deep learning-based CNN models and the size of feature vectors was reduced by feature selection algorithms. Obtained features were classified by traditional machine learning methods such as SVMs. The number of features to be selected was tested by the iterative increment method and the parameters with the highest accuracy rate were obtained. As a result, it was seen that healthy and infected individuals were detected in 3 classes with an accuracy rate of 98.78%. In addition, the confusion matrix, precision, recall values, and F1 score of the obtained model are also given.