Feature Vector Space Research Articles

Identification of mineralogical ore varieties using ultrasonic measurement results

Purpose. To improve the measurement and information base of ultrasonic measurements rock characteristics to assess their mineralogical varieties. It is proposed to use a combination of measurement results of the acoustic quality factor of the test sample in relation to longitudinal and transverse ultrasonic waves, as well as the characteristic coefficient based on the dispersion and the average amplitude value of the received signal, for fuzzy identification of mineralogical and technological varieties of iron ore. Methods. As elastic waves propagate through the rock mass, they undergo attenuation due to absorption and dissipation of ultrasonic signal energy. The degree of attenuation, as well as the wave propagation velocity, is dependent on the physical-mechanical and chemical-mineralogical properties of the medium through which they travel. In this paper, we analyze a rock characterized by a complex structure comprising ore inclusions and surrounding matrix, each of which differs in its physical-mechanical and chemical-mineralogical properties. In particular, in iron ore samples, the distribution of mineral grains and aggregates exhibits significant heterogeneity in terms of both amount and size. Findings. An iterative method of fuzzy identification of mineralogical-technological iron ore varieties, based on the analysis of their properties in vector space of features, allows, by minimizing the sums of weighted distances between the analyzed and reference values of ultrasonic measurement results, to attribute them with a certain degree of belonging to the main technological types of ores mined at the deposit, and define them as magnetite quartzite with a confidence probability of 0.93. Originality. As an information base for identification of mineralogical iron ore varieties, the results of measuring the velocity and attenuation of longitudinal and transverse ultrasonic waves of appropriate frequency are used, on the basis of which the acoustic quality factor of the rock sample is calculated, as well as the characteristic parameter S, which is determined by the dispersion and average values of the received ultrasonic signal intensity, which has traveled a certain distance in the studied environment. Practical implications. The results of tests and practical approbation of the method for identifying mineralogical iron ore varieties based on the data of ultrasonic well logging testify to its high efficiency, which allows recommending the developed scientific-technical solutions for wide industrial application at mining enterprises.

Hidden code vulnerability detection: A study of the Graph-BiLSTM algorithm

Context:The accelerated growth of the Internet and the advent of artificial intelligence have led to a heightened interdependence of open source products, which has in turn resulted in a rise in the frequency of security incidents. Consequently, the cost-effective, fast and efficient detection of hidden code vulnerabilities in open source software products has become an urgent challenge for both academic and engineering communities. Objectives:In response to this pressing need, a novel and efficient code vulnerability detection model has been proposed: the Graph-Bi-Directional Long Short-Term Memory Network Algorithm (Graph-BiLSTM). The algorithm is designed to enable the detection of vulnerabilities in Github’s code commit records on a large scale, at low cost and in an efficient manner. Methods:In order to extract the most effective code vulnerability features, state-of-the-art vulnerability datasets were compared in order to identify the optimal training dataset. Initially, the Joern tool was employed to transform function-level code blocks into Code Property Graphs (CPGs). Thereafter, structural features (degree centrality, Katz centrality, and closeness centrality) of these CPGs were computed and combined with the embedding features of the node sequences to form a two-dimensional feature vector space for the function-level code blocks. Subsequently, the BiLSTM network algorithm was employed for the automated extraction and iterative model training of a substantial number of vulnerability code samples. Finally, the trained algorithmic model was applied to code commit records of open-source software products on GitHub, achieving effective detection of hidden code vulnerabilities. Conclusion:Experimental results indicate that the PrimeVul dataset represents the most optimal resource for vulnerability detection. Moreover, the Graph-BiLSTM model demonstrated superior performance in terms of accuracy, training cost, and inference time when compared to state-of-the-art algorithms for the detection of vulnerabilities in open-source software code on GitHub. This highlights the significant value of the model for engineering applications.

GR‐Former: Graph‐reinforcement transformer for skeleton‐based driver action recognition

AbstractIn in‐vehicle driving scenarios, composite action recognition is crucial for improving safety and understanding the driver's intention. Due to spatial constraints and occlusion factors, the driver's range of motion is limited, thus resulting in similar action patterns that are difficult to differentiate. Additionally, collecting skeleton data that characterise the full human posture is difficult, posing additional challenges for action recognition. To address the problems, a novel Graph‐Reinforcement Transformer (GR‐Former) model is proposed. Using limited skeleton data as inputs, by introducing graph structure information to directionally reinforce the effect of the self‐attention mechanism, dynamically learning and aggregating features between joints at multiple levels, the authors’ model constructs a richer feature vector space, enhancing its expressiveness and recognition accuracy. Based on the Drive & Act dataset for composite action recognition, the authors’ work only applies human upper‐body skeleton data to achieve state‐of‐the‐art performance compared to existing methods. Using complete human skeleton data also has excellent recognition accuracy on the NTU RGB + D‐ and NTU RGB + D 120 dataset, demonstrating the great generalisability of the GR‐Former. Generally, the authors’ work provides a new and effective solution for driver action recognition in in‐vehicle scenarios.

Exploring The Impact of Stemming on Text Topic-Based Classification Accuracy

Text classification attempts to assign written texts to specific group types that share the same linguistic features. One class of features that have been widely employed for a wide range of classification tasks is lexical features. This study explores the impact of stemming on text classification using lexical features. To explore, this study is based on a corpus of thirty texts written by six authors with topics that focus on politics, history, science, prose, sport, and food. These texts are stemmed using a light stemming algorithm. In order to classify these texts according to the topic by means of lexical features, linear hierarchical clustering and non-linear clustering (SOM) is carried out on the stemmed and unstemmed texts. Although both clustering methods are able to classify texts by topic with two models produce accurate and stable results, the results suggest that the impact of a light stemming on the accuracy of text classification by topic is ineffectual. The accuracy is neither increased nor decreased on the stemmed texts, whereby the stemming algorithm helped reducing the dimensionality of feature vector space model.

Point-Sim: A Lightweight Network for 3D Point Cloud Classification

Analyzing point clouds with neural networks is a current research hotspot. In order to analyze the 3D geometric features of point clouds, most neural networks improve the network performance by adding local geometric operators and trainable parameters. However, deep learning usually requires a large amount of computational resources for training and inference, which poses challenges to hardware devices and energy consumption. Therefore, some researches have started to try to use a nonparametric approach to extract features. Point-NN combines nonparametric modules to build a nonparametric network for 3D point cloud analysis, and the nonparametric components include operations such as trigonometric embedding, farthest point sampling (FPS), k-nearest neighbor (k-NN), and pooling. However, Point-NN has some blindness in feature embedding using the trigonometric function during feature extraction. To eliminate this blindness as much as possible, we utilize a nonparametric energy function-based attention mechanism (ResSimAM). The embedded features are enhanced by calculating the energy of the features by the energy function, and then the ResSimAM is used to enhance the weights of the embedded features by the energy to enhance the features without adding any parameters to the original network; Point-NN needs to compute the similarity between each feature at the naive feature similarity matching stage; however, the magnitude difference of the features in vector space during the feature extraction stage may affect the final matching result. We use the Squash operation to squeeze the features. This nonlinear operation can make the features squeeze to a certain range without changing the original direction in the vector space, thus eliminating the effect of feature magnitude, and we can ultimately better complete the naive feature matching in the vector space. We inserted these modules into the network and build a nonparametric network, Point-Sim, which performs well in 3D classification tasks. Based on this, we extend the lightweight neural network Point-SimP by adding some trainable parameters for the point cloud classification task, which requires only 0.8 M parameters for high performance analysis. Experimental results demonstrate the effectiveness of our proposed algorithm in the point cloud shape classification task. The corresponding results on ModelNet40 and ScanObjectNN are 83.9% and 66.3% for 0 M parameters—without any training—and 93.3% and 86.6% for 0.8 M parameters. The Point-SimP reaches a test speed of 962 samples per second on the ModelNet40 dataset. The experimental results show that our proposed method effectively improves the performance on point cloud classification networks.

Personalized Federated Learning with Adaptive Feature Extraction and Category Prediction in Non-IID Datasets

Federated learning trains a neural network model using the client’s data to maintain the benefits of centralized model training while maintaining their privacy. However, if the client data are not independently and identically distributed (non-IID) because of different environments, the accuracy of the model may suffer from client drift during training owing to discrepancies in each client’s data. This study proposes a personalized federated learning algorithm based on the concept of multitask learning to divide each client model into two layers: a feature extraction layer and a category prediction layer. The feature extraction layer maps the input data to a low-dimensional feature vector space. Furthermore, the parameters of the neural network are aggregated with those of other clients using an adaptive method. The category prediction layer maps low-dimensional feature vectors to the label sample space, with its parameters remaining unaffected by other clients to maintain client uniqueness. The proposed personalized federated learning method produces faster learning model convergence rates and higher accuracy rates for the non-IID datasets in our experiments.

A bearing remaining life prediction method under variable operating conditions based on cross-transformer fusioning segmented data cleaning

Bearing remaining useful life (RUL) prediction research based on deep learning mostly emphasizes model performance and effective feature vectors, overlooking different densities of outlier distributions in vibration signals at varying degradation stages. Moreover, forecasting models focus on capturing cross-time dependencies, ignoring the dependencies between different variables. To solve these problems, this paper proposes an unsupervised segmented data cleaning algorithm and a RUL prediction framework adaptable to variable operating conditions. The method consists of four steps: (1) Multi-domain feature extraction and selection establish a feature vector space reflecting degradation trends. (2) Segmented data cleaning divides degradation stages, using different penalty factors for outlier cleaning. (3) Cleaned vibration signals undergo a second round of multidomain feature engineering and degradation-stage division. (4) A two-stage Cross-Transformer model is used for RUL prediction. The method proposed has been validated on the prognostics and health management (PHM) bearing degradation dataset. In the constant condition prediction task, the root mean square error (RMSE) and mean absolute error (MAE) were improved to 1.88 and 5.78, respectively. In the variable condition prediction task, the proposed method outperformed existing methods, with an improvement of 59.10 % in RMSE, demonstrating strong generalization performance and practical application value.

Automated building damage detection on digital imagery using machine learning

Purpose. To develop an automated method based on machine learning for accurate detection of features of a damaged building on digital imagery. Methodology. This article presents an approach that employs a combination of unsupervised machine learning techniques, specifically Principal Component Analysis (PCA), K-means clustering, and Density-Based Spatial Clustering of Applications with Noise (DBSCAN), to identify building damage resulting from military conflicts. The PCA method is utilized to identify principal vectors representing the directions of maximum variance in the data. Subsequently, the K-means method is applied to cluster the feature vector space, with the predefined number of clusters reflecting the number of principal vectors. Each cluster represents a group of similar blocks of image differences, which helps to identify significant features associated with fractures. Finally, the DBSCAN method is employed to identify areas where points with similar characteristics are located. Subsequently, a binary fracture mask is generated, with pixels exceeding the threshold being identified as fractures. Findings. The introduced methodology attains an accuracy rate of 98.13 %, surpassing the performance of conventional methods such as DBSCAN, PCA, and K-means. Furthermore, the method exhibits a recall of 82.38 %, signifying its ability to effectively detect a substantial proportion of positive examples. Precision of 58.54 % underscores the methodology’s capability to minimize false positives. The F1 Score of 70.90 % demonstrates a well-balanced performance between precision and recall. Originality. DBSCAN, PCA and K-means methods have been further developed in the context of automated detection of building destruction in aerospace images. This allows us to significantly increase the accuracy and efficiency of monitoring territories, including those affected by the consequences of military aggression. Practical value. The results obtained can be used to improve automated monitoring systems for urban development and can also serve as the basis for the development of effective strategies for the restoration and reconstruction of damaged infrastructure.

Identifying the Causes of Ship Collisions Accident Using Text Mining and Bayesian Networks

Under the backdrop of the robust growth of the global economy, the water transport industry is experiencing rapid development, resulting in an increase in ship collisions and a critical water traffic safety situation. This study uses text mining techniques to gather a corpus of data. The corpus includes human factors, ship factors, natural environmental factors, and management factors, which are used as target data to obtain a high-dimensional sparse original feature vector space set comprising eigenvalues and eigenvalue weight attributes. Chi-square statistics are utilised to reduce dimensionality, resulting in a final set of 33-dimensional text feature items that determine the causal factors of ship collision risk. Taking the four steps involved in the collision process as the primary focus, a Bayesian network structure for ship collision risk is constructed based on the “human-ship-environment-management” system. By incorporating existing ship collision accident/danger reports, conditional probability tables are computed for each node in the Bayesian network structure, enabling the modelling of ship collision risk. The model is validated through an example, revealing that, under relevant conditions, the probability of collision exceeds 90 %. This finding demonstrates the validity of the model and allows one to deduce the primary cause of ship collision accidents.

Generative and Discriminative Learning for Lung X-Ray Analysis Based on Probabilistic Component Analysis.

The paper presents a hybrid generative/discriminative classification method aimed at identifying abnormalities, such as cancer, in lung X-ray images. The proposed method involves a generative model that performs generative embedding in Probabilistic Component Analysis (PrCA). The primary goal of PrCA is to model co-existing information within a probabilistic framework, with the intent to locate the feature vector space for X-ray data based on a defined kernel structure. A kernel-based classifier, grounded in information-theoretic principles, was employed in this study. The performance of the proposed method is evaluated against nearest neighbour (NN) classifiers and support vector machine (SVM) classifiers, which use a diagonal covariance matrix and incorporate normal linear and non-linear kernels, respectively. The method is found to achieve superior accuracy, offering a viable solution to the class of problems presented. Accuracy rates achieved by the kernels in the NN and SVM models were 95.02% and 92.45%, respectively, suggesting the method's competitiveness with state-of-the-art approaches.

A study of innovations in legal governance with respect to the safety of artificial intelligence

Abstract This paper aims to promote the safe development of artificial intelligence and improve legal policies. Combined with the cluster analysis algorithm, it analyzes the safety risks as well as legal defects of artificial intelligence. The Laplace matrix is derived using the similarity matrix, and the feature vector space is constructed by analyzing the associated features of artificial intelligence safety. Combining the spectral clustering algorithm, legal assessment indexes for artificial intelligence safety were constructed. The modular metric value method is utilized to assess the clustering effect of laws on the safety of artificial intelligence. Analyzing the security risks of artificial intelligence, improved legal policies are proposed from the perspective of technology and privacy. The results show that the effect of improving privacy protection policy on privacy protection is 0.85, and the effect of clarifying subject rights is 0.9. The introduction of laws should consider social ethics, and the effect degree of ethical principles is 0.75. Clarifying subject rights can help avoid technological risks to a certain extent, and improving privacy protection policies can help protect users’ privacy.

A novel scheme based on modified hierarchical time-shift multi-scale amplitude-aware permutation entropy for rolling bearing condition assessment and fault recognition

To avoid production interruptions and equipment damage caused by rolling bearing failure, this study presents a novel diagnosis scheme applicable to both condition monitoring and fault recognition. In detail, to fully exploit feature information, a modified Hierarchical Time-Shift Multi-scale Amplitude-Aware Permutation Entropy (MHTSMAAPE) is put forward to map the raw vibration signal into a high-dimensional vector space of features. This method is developed from the AAPE algorithm and integrates the time-shift procedure and modified hierarchical analysis, extracting more feature details while ensuring entropy stability. Afterwards, the averaged value of the feature vector is regarded as an indicator to assess the condition of rolling bearing, this indicator exceeds a specific value indicates the rolling bearing steps into faulty. Then, the Unsupervised Discriminative Feature Selection (UDFS) algorithm is first introduced to directly pick fault-sensitive features from the vector space, relieving the calculation burden of the fault recognition task. The fault identification algorithm is based on Harris Hawks Optimization-Joint Opposite Selection optimized for Support Vector Machine (HHO-JOS-SVM) to improve classification accuracy. Finally, the proposed scheme fused by MHTSMAAPE, UDFS, and HHO-JOS-SVM is validated in two different cases with four common evaluation metrics, respectively. The experimental results demonstrate its reliability and robustness in rolling bearing condition assessment and fault recognition.

Few-shot classification with multisemantic information fusion network

Metric-based methods aim to predict class labels by computing the similarity between samples using distance functions, which is the mainstream approach to few-shot learning. However, the limited representational space of feature vectors and appearance variations among congenetic samples still present challenges. We propose a Multisemantic Information Fusion Network (MIFN) to address these problems. A Lower-level Feature generator (LF-generator), which is an unsupervised module, adaptively activates high-response regions of objects to introduce discriminative semantic details. Meanwhile, a Higher-level Feature extractor (HF-extractor) learns global semantic information with human cognition to minimise the impact of appearance variations. We integrate the coarse outputs of these two modules, which complement each other to jointly promote more precise predictions. Furthermore, considering the importance of prototypes, we redefine the sampling strategy of the triplet loss and utilise it as an auxiliary loss to sharpen the decision boundary at the prototype level, facilitating subsequent classification. Our experimental results demonstrate the competitiveness of our approach in both general few-shot classification (mini-ImageNet and tiered-ImageNet) and cross-domain problems (CUB, Caltech-101, Stanford-dogs, and Stanford-cars) with minimal bells and whistles.

Facial Emotion Recognition for Photo and Video Surveillance Based on Machine Learning and Visual Analytics

Modern video surveillance systems mainly rely on human operators to monitor and interpret the behavior of individuals in real time, which may lead to severe delays in responding to an emergency. Therefore, there is a need for continued research into the designing of interpretable and more transparent emotion recognition models that can effectively detect emotions in safety video surveillance systems. This study proposes a novel technique incorporating a straightforward model for detecting sudden changes in a person’s emotional state using low-resolution photos and video frames from surveillance cameras. The proposed technique includes a method of the geometric interpretation of facial areas to extract features of facial expression, the method of hyperplane classification for identifying emotional states in the feature vector space, and the principles of visual analytics and “human in the loop” to obtain transparent and interpretable classifiers. The experimental testing using the developed software prototype validates the scientific claims of the proposed technique. Its implementation improves the reliability of abnormal behavior detection via facial expressions by 0.91–2.20%, depending on different emotions and environmental conditions. Moreover, it decreases the error probability in identifying sudden emotional shifts by 0.23–2.21% compared to existing counterparts. Future research will aim to improve the approach quantitatively and address the limitations discussed in this paper.

A neural kernel method for capturing multiscale high-dimensional micromorphic plasticity of materials with internal structures

This paper introduces a neural kernel method to generate machine learning plasticity models for micropolar and micromorphic materials that lack material symmetry and have internal structures. Since these complex materials often require higher-dimensional parametric space to be precisely characterized, we introduce a representation learning step where we first learn a feature vector space isomorphic to a finite-dimensional subspace of the original parametric function space from the augmented labeled data expanded from the narrow band of the yield data. This approach simplifies the data augmentation step and enables us to constitute the high-dimensional yield surface in a feature space spanned by the feature kernels. In the numerical examples, we first verified the implementations with data generated from known models, then tested the capacity of the models to discover feature spaces from meso-scale simulation data generated from representative elementary volume (RVE) of heterogeneous materials with internal structures. The neural kernel plasticity model and other alternative machine learning approaches are compared in a computational homogenization problem for layered geomaterials. The results indicate that the neural kernel feature space may lead to more robust forward predictions against sparse and high-dimensional data.

Opinion Mining with Density Forests

In this paper, we propose a new approach for opinion mining with density-based forests. We apply Density-Based Spatial Clustering of Applications with Noise (DBSCAN) to identify clusters of data points in a space of feature vectors that are important features of hotel and restaurant reviews, and then use the clusters to construct random forests to classify whether the opinions expressed about features in the reviews are positive or negative. Our experiment uses two standard datasets of hotel and restaurant reviews in two different scenarios. The experimental results show the effectiveness of our proposed

Sparse representation learning using [formula omitted] compressed sensing and rank-revealing QR factorization

Compressive sensing can be conceptualized for a classification problem in the context of representation learning, which is frequently applied for signal reconstruction via a few measurements. A novel compressive sensing-based approach is presented to substantially reduce the number of required samples for classification. This technique considers the pixels of image as sensors and it seeks to identify the most optimum sensors as decision variables in feature space. In this study, Spatial sensor locations are acquired through learning to identify the regions within an image where the most discriminative information for classification is embedded. ℓ1−2 minimization, as nonconvex but Lipschitz continuous, is solved to obtain the least non-zero elements of the full measurement vector to completely reconstruct the discriminative vector in feature space. Optimal sensors are localized from the training-set and subsequent test images are categorized based on learned sensors. The algorithm consists of three primary components: sparse minimization, feature space, and discrimination vector. ℓ1−2 minimization, rank-revealing QR factorization (RRQR) and SVM are considered for enhanced sparsity exploitation, feature space extraction and discrimination vector acquiring, respectively. The proposed method is evaluated on four different experiments and is compared against a state-of-art technique. The results demonstrate the superiority of proposed method to the compared method.

Evading Machine-Learning-Based Android Malware Detector for IoT Devices

Securing the Android gadgets from malware activities is a major concern in Internet of Things, and the drastic rise in adversarial attacks makes this problem more challenging. Existing Android adversarial techniques focus on the feature vector space manipulations through complex deep learning techniques, and the operational integrity is also not experimented. This work implements two evasion attack scenarios on the feature vector of Android apks and creates functionality preserved Android malware. The malware samples are injected with features that are absent in malicious samples, but found in legitimate applications. The sample thus formed will be identical in functionality, besides having statistical dissimilarity. One of the attack scenarios implemented is through feature similarity using Euclidean distance (ED), which measures between malware and benign samples. The other version of attack is by generating variants through particle swarm optimization (PSO). We find that our evasion algorithms are not highly biased nor complex, making them easier to train and understand compared to generative adversarial neural networks. The experiments were carried out on real-world Android applications from AndroZoo and AMD. We could successfully achieve the highest performance of 100% evasion rate with the PSO algorithm and 89.6% evasion rate with the ED algorithm with the lowest computational complexity compared to machine learning or deep learning mechanisms.

A rumor heat prediction model based on rumor and anti-rumor multiple messages and knowledge representation

In the field of online social networks, the effective prediction of group behavior is the key to fitting the trajectory of rumor topic propagation. Considering the heterogeneity and complexity of a rumor topic propagation network, this paper proposes a rumor heat prediction model based on multiple rumor and anti-rumor messages and knowledge representation. Firstly, aiming at the dynamic interaction of multiple messages under the rumor topic network, this paper introduces evolutionary game theory to quantify the influence of multiple messages on group behavior to truly reflect the relationship structure of the rumor topic spreading network. Secondly, to solve the difficulties in mining the complex features of multisource entities, knowledge representation is considered the theoretical basis to map the heterogeneous topic network to the feature vector space and learn the low-rank vector representation of the complex features. Finally, considering the time-sensitive nature of rumor propagation and the advantages of graph neural networks in processing non-Euclidean structured data, the dynamic rumor topic heat prediction model called K-GraphSAGE, which is based on group behavior, is proposed. The experiment results show that this method can not only effectively predict the group behavior and the heat of the rumor topics, but also unearth the mechanism behind the spreading law of the rumor topics.

Face image de‐identification by feature space adversarial perturbation

SummaryPrivacy leakage in images attracts increasing concerns these days, as photos uploaded to large social platforms are usually not processed by proper privacy protection mechanisms. Moreover, with advanced artificial intelligence (AI) tools such as deep neural network (DNN), an adversary can detect people's identities and collect other sensitive personal information from images at an unprecedented scale. In this paper, we introduce a novel face image de‐identification framework using adversarial perturbations in the feature space. Manipulating the feature space vector ensures the good transferability of our framework. Moreover, the proposed feature space adversarial perturbation generation algorithm can successfully protect the identity‐related information while ensuring the other attributes remain similar. Finally, we conduct extensive experiments on two face image datasets to evaluate the performance of the proposed method. Our results show that the proposed method can generate real‐looking privacy‐preserving images efficiently. Although our framework has only been tested on two real‐life face image datasets, it can be easily extended to other types of images.