Search

Abstract: In this paper, we describe the development of a system for evaluating an specific quality characteristic of recycled paper sheets using techniques of image analysis and pattern recognition. We call Bumpiness the phenomenon of interest, which is new in the literature on paper quality. This phenomenon is characterized by the appearance of macroscopic undulations on the paper sheet surface that may emerge shortly or some time after its production. We explore the detection and measurement of this defect by means of computer vision and statistical pattern recognition techniques that may allow early detection at the production site. Our goal is to give an scalar continuous measure of Bumpiness. We propose features computed from Gabor filter banks (GFB) and discrete wavelet transforms (DWT) for the characterization of paper sheet surface Bumpiness in recycled paper images. The starting point is to state the problem as a classification of the paper sheet images into two classes: low and high Bumpiness. In this setting we obtain, with both proposed texture modelling approaches (GFB and DWT), classification accuracies comparable to the agreement between human observers. The best performance is obtained using DWT features. Finally, we propose as the scalar index of Bumpines the fisher discriminant analysis (FDA) function defined on the space of the best features for the classification task. We perform an innovative validation process of this Bumpiness index, based on the ordering of random pairs of images, obtaining a very high agreement with the human observers. [Copyright &y& Elsevier]

Feature selection is a crucial aspect of data preprocessing because of the significant effect of redundant features on classification performance and the extensive computational resources required. Evolutionary algorithm-based feature-selection methods have shown remarkable results in determining the optimal feature subset. To further enhance classification performance, this paper proposes a novel multi-population differential evolution approach for feature selection with mutual information ranking (MI-MPODE). Firstly, population preprocessing guided by mutual information is employed to reduce the dimensionality of the initial feature space. Then, the feature subset obtained from mutual information serves as the initial population for MI-MPODE. MI-MPODE incorporates a novel multi-population information-sharing mechanism, with common individuals from three layers contributing to inter-subpopulation information sharing. Additionally, an individual enhancement strategy is proposed to handle the variations of individuals in the population, and a Lens imaging opposition-based learning method is adopted to improve the algorithm's optimization capability. MI-MPODE is compared with several state-of-the-art evolutionary algorithm-based feature selection methods through experimental comparisons. The results show that MI-MPODE outperforms all comparison algorithms on more than half of the datasets, with a significant reduction in the number of features used, demonstrating a significant advantage over competitors. • This paper proposes a novel multi-population information sharing mechanism. • MI technology to optimize the initial population. • The individual stratification method and the LensOBL strategy improve the algorithm performance. • The performance of MI-MPODE is better than competitors on many datasets. [ABSTRACT FROM AUTHOR]

This paper evaluates five strategies for mitigating imbalanced data: oversampling, undersampling, ensemble methods, specialized algorithms, class weight adjustments, plus a no-mitigation approach referred to as the baseline. These strategies were tested on 58 real-life binary imbalanced datasets with imbalance rates ranging from 3 to 120. We conducted a comparative analysis of 10 under-sampling algorithms, 5 over-sampling algorithms, 2 ensemble methods, and 3 specialized algorithms across eight different performance metrics: accuracy, area under the ROC curve (AUC-ROC), balanced accuracy, F1-measure, G-mean, Matthew's correlation coefficient, precision, and recall. Additionally, we assessed the five strategies plus the baseline on altered datasets, derived from real-life data, with both low (3) and high (100 or 300) imbalance ratios (IR). This paper tries to assist a practitioner in selecting a small set of algorithms within one or few of the strategies for testing for a specific imbalanced problem. The principal finding indicates that the effectiveness of each strategy significantly varies depending on the metric used. The paper also examines a selection of newer algorithms within the categories of specialized algorithms, oversampling, and ensemble methods. The findings suggest that the current hierarchy of best-performing strategies for each metric is unlikely to change with the introduction of newer algorithms. • This paper tests imbalanced algorithms grouped under five strategies. • The paper tests 20 algorithms on 53 imbalanced data sets, testing 6 different metrics of quality. • The effectiveness of each strategy varies significantly depending on the evaluation metric employed. • The paper suggests that the ranking of the strategies remains stable with the introduction of new algorithms within each strategy. [ABSTRACT FROM AUTHOR]

• First comprehensive, systematic review of deep learning in radiology for lung cancer. • Review 153 papers for segmentation, survival prediction, subtype classification, etc. • Categorize state-of-the-art models on different imaging modalities (PET, CT, PET/CT). • Present the best models across each image modality and lung cancer-related task. • Review current challenges, future directions for efficient lung cancer investigation. This study presents a comprehensive systematic review focusing on the applications of deep learning techniques in lung cancer radiomics. Through a rigorous screening process of 589 scientific publications following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines, we selected 153 papers for an in-depth analysis. These papers were categorized based on imaging modality, deep learning model type, and practical applications in lung cancer, such as detection and survival prediction. We specifically emphasized deep learning models and examined their strengths and limitations for each application and imaging modality. Furthermore, we identified potential limitations within the field and proposed future research directions. This study serves as a pioneering resource, being the first comprehensive and systematic review of deep learning techniques, specifically in the context of lung cancer-related applications. Our primary objective was to provide a reference for future research, encouraging the advancement of deep learning techniques in the diagnosis and treatment of lung cancer. By suggesting the most effective deep learning tools for specific application areas, we offer a benchmark for future studies. In summary, this study consolidates and expands existing knowledge on deep learning and radiomics applications in lung cancer. It provides a foundation for further research and serves as a guide for developing and evaluating deep learning models in lung cancer-related applications. [ABSTRACT FROM AUTHOR]

Abstract: In this paper it is introduced a recognition method for Mexican banknotes by using artificial vision. It is shown that the Mexican banknotes can be classified by extracting their color and texture features, with the RGB space and the Local Binary Patterns, respectively. We show the classification results performed with the current Mexican banknotes. We state that the proposed method can be applied to recognize banknotes of other countries which employ colors to distinguish the denominations. [Copyright &y& Elsevier]

Relying solely on sentiment polarity to express users' true attitudes towards rumor-refuting information is imprecise, and methods that ignore specific user groups and their influence on refutation will result in ineffective spread of refutation information. In this paper, we investigate three questions: (1) How to measure the users' attitudes of rumor-refuting information? (2) How can users be more efficiently classified based on their attitudes and behaviors? (3) What are the group features of different crowds? Accordingly, we introduce an innovative classification scheme based on users' attitudes and interactive behaviors. Subsequently, Natural language processing (NLP) techniques are applied to quantify the sentiment tendency and recent interests of users. Believability as a metric for quantifying users' attitudes is measured by the proposed RoBERTa-BiLSTM model. Then, five machine learning classification models are established, among which Extreme Gradient Boosting (XGBoost) model has the best performance. Lastly, based on XGBoost and Target Group Index (TGI), four groups are comprehensively evaluated and user profiles are constructed for distinct groups: rumor refuters, rumor spreaders, anti-rumor inactivists and pro-rumor inactivists. The results reveal that the distribution of Believability shows significant polarization among users and there are significant feature differences between various groups. This work makes a notable contribution to refutation by advancing beyond traditional sentiment analysis to precise measurement of users' attitudes toward rumor-refuting content. By introducing the novel classification scheme, this paper is the first of its kind for applying Believability of rumor refutation and offering a more nuanced understanding of users' attitudes as a differentiator for crowd classification. It advances the understanding and development of crowd classification and holds the potential to significantly impact the realm of precise rumor governance. [ABSTRACT FROM AUTHOR]

Local binary pattern (LBP) and many variants adopt the difference information to encode all the neighborhood binarization, demonstrating excellent ability to distinguish images. However, these methods are very sensitive to reverse gray changes. How to improve the robustness of reverse gray changes and maintain excellent classification ability has become a problem to be solved. In this paper, an image captioning method called multiscale image captioning method based on wavelet transform is proposed (WMLBP). This method uses stationary wavelet transform to remove the redundant information of the image and extract features from the frequency domain. This paper also proposes an innovative coding scheme, which designs three descriptors, and uses multi-scale hierarchical thresholds and complementary information to enhance the ability of image feature description. Finally, the image features were described by combining all the descriptors and using the multi-scale and multi-resolution cross-scale joint representation. Tests across multiple databases have demonstrated that the preprocessing of this method can significantly enhance classification capabilities. Compared to other algorithms, it has a higher accuracy rate and faster computational speed. Experiments demonstrate that the method exhibits strong robustness against grayscale inversion changes (both linear and nonlinear) and image rotation, while also demonstrating excellent classification performance. It holds significant importance for addressing the challenges encountered in image classification. Our code is available at: https://github.com/Dawei-W/WMLBP/tree/master. [ABSTRACT FROM AUTHOR]

Representation based classification (RC) algorithms have been successfully applied to pattern classification. However, most existing RC algorithms are not robust to bad training samples, since they ignore learning more discriminative dictionary atoms and instead directly use training samples as dictionary atoms. In addition, in order to improve expression ability and recognition accuracy, RC algorithms often need to expand the number of dictionary atoms, resulting in a surge in storage and computing costs. To obtain a more discriminative and compact dictionary, this study proposes discriminative dictionary learning for nonnegative representation based classification (DDLNRC). Specifically, in the DDLNRC, this paper utilizes nonnegative constraint to obtain a nonnegative representation for each training sample on the dictionary. In the dictionary learning stage, for a training sample, the DDLNRC minimizes the intra-class reconstruction error of the training sample and simultaneously enlarges the distance between the training sample and the atom that has the most influence on inter-class reconstruction error. Experiments demonstrate the effectiveness of the DDLNRC. Combined with the deep neural network features, it also can achieve higher accuracy than using the Softmax. • This paper proposes a DDLNRC for pattern recognition. • The DDLNRC can minimize intra-class reconstruction error. • The DDLNRC also can enlarge the inter-class reconstruction error. • By DDLNRC, a discriminative and compact dictionary can be obtained. • Experiments demonstrate the effectiveness of the DDLNRC on pattern recognition. [ABSTRACT FROM AUTHOR]

Hierarchical text classification (HTC) aims to classify a text into multiple categories organized in a hierarchical structure. The state-of-the-art HTC methods usually employ graph networks, where label graphs are constructed and label representation is learned to interact with text representations for classification. In general, label graphs are built on the intrinsic label hierarchy, label semantic similarity, or label co-occurrence. Such graphs have been proven to be effective, but they only exploit knowledge from training data or simple label descriptions, without considering the vast external knowledge in the open sources. Actually, external knowledge from open sources could bring in complementary information to enhance the label graph's representation power. Motivated by the above considerations, we explore the use of external knowledge for improving HTC in this paper. We categorize knowledge into micro-knowledge and macro-knowledge, which are defined as the fundamental concepts related to a single class label and the correlations among class labels, respectively. For tailor-made incorporation of the two types of knowledge into representation learning and classification, we propose Adaptive Micro- and Macro-Knowledge Incorporation for Hierarchical Text Classification (AMKI-HTC) model in this paper. The micro-knowledge incorporation helps capture class-relevant keywords in the text and hence produce discriminative representations, while the macro-knowledge incorporation improves the accuracy of label graphs. Finally, a confidence maximization fusion strategy is developed for adaptive aggregation of multi-view features. Extensive experiments on three benchmark HTC datasets demonstrate that AMKI-HTC consistently outperforms state-of-the-art models. • Conceptualize extensive external knowledge as micro-knowledge and macro-knowledge. • Aggregate micro- and macro-knowledge into original text features adaptively. • Construct text-knowledge graphs to emphasize discriminant information. • Simplify hierarchy-GCN as SH-GCN with improved performance. • Adaptively fuse features from multiple label graphs by confidence maximization. [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]

• Pinch force strength is accurately predicted using sEMG. • Different pinch movements are accurately classified using sEMG. • Similarity of EMG features is evaluated. • An EMG feature-to-image method is proposed. The movement of intelligent EMG-driven prosthesis mainly relies on the synergy of different fingers to achieve function of grasping objects. The paper proposes a novel scheme for force prediction and movement classification about pinch between different fingers based on surface electromyography (sEMG) using machine learning. The pinch force and sEMG signals are recorded synchronously by a data acquisition device. Eight features are extracted, which are proven to have better performance in the estimation of sEMG-to-force. We present a novel feature selection method that uses the one-dimensional time series similarity assessment based on Manhattan distance to eliminate the repetitive information between features. Three optimal features carrying less repetitive information are retained. Seven machine learning algorithms are used to predict force strength. The results show that the Long Short Term Memory (LSTM) has the best performance of force prediction, achieving a R 2 of 0.9517 and RMSE of 3.2723. The paper proposes a novel method of converting EMG feature sequences to the normalized gray image in order to classify the finger movement. Five classifiers based on image feature extraction and the Convolutional Neural Network (CNN) are developed respectively. The experimental results indicate that the CNN performs best, achieving an accuracy of 97.66%. In this way, it not only realizes the accurate force prediction, but also realizes the movement classification between different fingers. The proposed methodology has the potential to realize simultaneous force and movement control of prosthetic hand. [ABSTRACT FROM AUTHOR]

Human Activity Recognition (HAR) is the task to automatically analyze and recognize human body gestures or actions. HAR using time-series multi-modal sensory data is a challenging and important task in the field of machine learning and feature engineering due to its increasing demands in numerous real-world applications such as healthcare, sports and surveillance. Numerous daily wearable devices e.g., smartphones, smartwatches, and smart glasses can be used to collect and analyze the human activities on an unprecedented scale. This paper presents a generic framework to recognize the different human activities using continuous time-series multimodal sensory data of these smart gadgets. The proposed framework follows the channel of Bag-of-Features which consists of four steps: (i) Data acquisition and pre-processing, (ii) codebook computation, (iii) feature encoding, and (iv) classification. Each step in the framework plays a significant role to generate an appropriate feature representation of raw sensory data for efficient activity recognition. In the first step, we employed a simple overlapped-window sampling approach to segment the continuous time-series sensory data to make it suitable for activity recognition. Secondly, we build a codebook using k-means clustering algorithm to group the similar sub-sequences. The center of each group is known as codeword and we assume that it represents a specific movement in the activity sequence. The third step consists of feature encoding which transform the raw sensory data of activity sequence into its respective high-level representation for the classification. Specifically, we presented three reconstruction-based encoding techniques to encode sensory data, namely: Sparse Coding, Local Coordinate Coding, and Locality-constrained Linear Coding. The segmented activity sub-sequences are transformed to high-level representation using these techniques and earlier computed codebook. Finally, the encoded features are classified using a simple Random Forest classifier. The proposed HAR framework using three different encoding techniques is evaluated on three large benchmark datasets: UniMiB-SHAR dataset, MHEALTH dataset and WISDM dataset and their results are compared with most recent state-of-the-art techniques. It outperforms the existing techniques and achieves 98.39%, 99.5%, and 99.4% recognition scores on UniMiB-SHAR dataset, MHEALTH dataset, and WISDM dataset respectively. The excellent recognition results and computational analysis confirms the effectiveness and efficiency of the proposed framework. • A generic framework to analyze human activities using multimodal sensory data. • A comprehensive and systematic assessment of existing approaches. • An extensive evaluation of different reconstruction-based encoding methods. • Experimental evaluation of the encoding methods on three large benchmark datasets. [ABSTRACT FROM AUTHOR]

Convolutional neural networks play a vital role in image classification, with attention mechanisms enhancing discriminability on large datasets like ImageNet. However, challenges persist in optimizing performance for smaller or moderate real-world datasets due to limited data availability. There is a deficiency in effectively leveraging both channel and spatial attention for enhanced effectiveness in such scenarios. Although predefined filters offer advantages, their integration with attention mechanisms for complementary strength remains under-explored. In this paper, we propose the Frequency Regulated Channel-Spatial Attention (FReCSA) module to address this challenge by leveraging the power of channel attention and spatial attention. Four subsets and the complete ImageNet dataset, along with five additional datasets are used to evaluate FReCSA in our experiments. Integrating the FReCSA module into ResNet50 significantly enhances the top-1 accuracy, which is demonstrated by a 10.13% increase over the second-best on the ImageNet-40 dataset. On the other hand, our FReCSA module introduces minimal computational and parameter overhead to the deep network in terms of model size and computational operations, which makes FReCSA a good choice for learning tasks. The source code of this work is available at https://github.com/CoVIS-UNT/FReCSA. • A deep network module to address the problem of limited training examples. • A frequency-modulated attention integrating predefined filters. • The module can be incorporated into the most popular deep networks. • Experiments demonstrate an improvement of top-1 accuracy by 10.13%. [ABSTRACT FROM AUTHOR]

Deep subspace clustering (DSC) methods based on deep autoencoder (DAE) and self-expression layers have achieved impressive performance. However, traditional DSC method often loses useful information in the feature extraction of DAE, leading to incomplete learning of the self-expression coefficient matrix. Besides, existing deep subspace clustering methods always extract feature information from a single dimension, ignoring the potential information that may exist in the self-expression coefficient matrix across multiple dimensions. In this paper, we propose a novel method called multi-dimensional weighted deep subspace clustering based on feature classification (MWDSC). Specifically, we use the multivariate feature classification module (MFC) to learn the potential information of multiple local feature matrix for local pre-clustering. To further improve the clustering performance, we propose a self-flip weighted fusion strategy (SWF). The SWF strategy obtains a self-expression coefficient matrix with rich features through self-flipping weighting, thereby aiding in achieving global optimal solutions. Extensive experiments conducted on five benchmark datasets validate the superiority and effectiveness of our proposed method over other state-of-the-art approaches. • A new feature extraction module based on multi-channel is proposed. • A self-flip weighted fusion strategy highlights feature info across dimensions. • We apply the designed deep subspace clustering to experiment on five datasets. [ABSTRACT FROM AUTHOR]

The multilabel twin support vector machine (MLTSVM) has been widely applied to multilabel classification fields because of its excellent classification performance, but it has the following three disadvantages. (a) In practical classification tasks, due to human negligence or some objective factors, there are many data with noise labels. Noisy label data have a negative effect on generating classification hyperplanes. However, MLTSVM cannot effectively reduce the negative effects on noisy label data. (b) Each label has its own specific features. However, for each label, MLTSVM can only use the same feature representation for classification. Therefore, it cannot obtain the optimal classification results. (c) In multilabel classification problems, there is a general correlation among the labels. However, when MLTSVM constructs a classification hyperplane for each label, it ignores the label correlation information. To address the above disadvantages, in this paper, a novel algorithm is proposed the called intuitionistic fuzzy least squares MLTSVM for noisy label data using label-specific features and local label correlation (SFLLC-IFLSMLTSVM). First, SFLLC-IFLSMLTSVM constructs an intuitionistic fuzzy set for each label; second, SFLLC-IFLSMLTSVM exploits the membership and nonmembership degrees in the intuitionistic fuzzy set to filter out the samples with noise labels and then performs clustering analysis to extract the label-specific features and the structure information of samples; finally, SFLLC-IFLSMLTSVM mines the local label correlation information in the process of constructing classification hyperplanes and exploits the membership and nonmembership degrees to suppress the negative effects of noise labels. Extensive comparative experiments on multilabel datasets show that SFLLC-IFLSMLTSVM can efficiently handle the multilabel classification problems with noise labels. • Use intuitionistic fuzzy membership to filter out samples with noise labels. • Extract specific-label features and structural information of samples. • Use local label correlation information to construct classification hyperplanes. • Use intuitionistic fuzzy membership to suppress the negative effects of noise labels. [ABSTRACT FROM AUTHOR]

As a promising model compression technique, knowledge distillation aims to supervise the training of small networks with advanced knowledge from large networks to improve the performance of small networks. However, the performance improvement of student models is generally limited by the capacity gap between the teacher and student models. Previous approaches typically aim to have students mimic teacher features, which may lead to suboptimal results. In this paper, we investigate the effect of introducing random noise in intermediate features on knowledge distillation. The noise helps students to focus on learning key information and thus improves feature representation. Specifically, we propose Noisy Feature Distillation (NFD), which adds noise to student features and uses a convolutional block to reconstruct features under the guidance of teacher features. In this way, we create a noisy environment that allows the student and teacher networks to focus on objects of interest in different tasks, thereby improving the robustness of the knowledge and its transfer process. In addition, we introduce the spatial attention of teacher features to modulate the noise distribution and guide students to focus on key pixels, further improving student performance. Extensive experiments show that our strategy outperforms other methods in various tasks such as classification, object detection, instance segmentation, and semantic segmentation. Significant performance gains are achieved in both homogeneous and heterogeneous distillation. And our strategy excels in small object detection. • We explore the impact of random noise on the performance of distillation. • Appropriate noise helps the student features gain stronger representation power. • We propose a knowledge distillation method based on noisy feature reconstruction. • Our method is applicable to various tasks, e.g., classification and dense prediction. • Our method yields excellent distillation results, especially in heterogeneous cases. [ABSTRACT FROM AUTHOR]

Semi-supervised classification aims to leverage a small amount of labeled data for learning tasks. Multi-view semi-supervised classification has attracted widespread attention because it can exploit multi-view data to optimize the classification performance. However, its methods are often ineffective when facing extremely limited labeled samples. In this paper, we propose a novel multi-view semi-supervised classification model via auto-weighted submarkov random walk. The proposed method can utilize similar nodes, spread information among nodes on graphs and exploit multi-view data with less labeled information. Accordingly, it enables an effective exploitation of both a small number of labeled data and a large amount of unlabeled data by connecting them to designed auxiliary nodes. Furthermore, an ideal weight on the Hellinger distance is allocated to each view data for obtaining a global label indicator matrix, which is expected to be robust to imbalanced classes. Compared with existing state-of-the-art methods, extensive experiments on six widely used datasets are conducted to verify the superiority of the proposed method. • Introduce submarkov random walk to semi-supervised classification. • A closed-form solution is applied to handle semi-supervised classification. • Ideal weights are automatically designed for better label indicator. [ABSTRACT FROM AUTHOR]

In the field of data mining, outlier detection is an important and challenging task. This paper focuses on studying the impacts of outliers on the performance of models learned from large-scale datasets while it is unknown whether a data point is outlier or not. In the proposed approach, a fuzzy c-means clustering algorithm is applied and outliers are defined as those that have greater membership values in the cluster and are located further away from the cluster centroid. Ideally, the sample with a higher membership value should be located closer to the cluster centroid. In this context, we calculate the weight of each sample using the AdaBoost algorithm, where a weight determines the representativeness of each sample within the data distribution. Additionally, in this study, the impact of weighted loss functions in different situations are discussed in detail. At last, our method is evaluated on 12 UCI datasets and the accuracy of our method is greater than 95% on some datasets, such as banknote 99.99%, biodeg 99.01%, optdigits 99.19%, and letters 97.37%. The experimental results show the efficiency and effectiveness of the proposed approach. • Outlier impact is investigated in a classification context. • Outliers are identified as having higher membership and found outside cluster center. • Different weighted loss functions are also discussed in detail in various situations. • The efficiency and effectiveness of the proposed method are verified experimentally. [ABSTRACT FROM AUTHOR]

• RFM generates multiple feature subsets by random feature mapping for stabilizing. • RFM enhances classification effect by fusing results of multiple feature subsets. • RFM assigns optimal weights to different weak classifiers by AdaBoost algorithm. • RFM processes datasets of different dimensions and distribution adaptively. The feature mapping method can improve data separability, enhance data representation ability, and reduce data processing complexity. However, on the one hand, the existing feature mapping methods have difficulty processing datasets of different dimensions and distributions adaptively, limiting the scope of application; on the other hand, a single feature mapping method has the problem of instability and poor generalization ability, weakening the classification ability of subsequent classifiers. This paper proposes a random feature mapping method based on the AdaBoost algorithm and results fusion to enhance classification performance. The method adopts horizontal expansion, fine-tuning weights through sparse autoencoders, and uses input-mapped features as feature nodes to generate multiple feature subsets for increasing stability. After training weak classifiers on each multiple feature subset, the weights of classifiers are adjusted adaptively by the Adaboost ensemble algorithm. Finally, the method fuses weak classifiers twice to enhance classification performance, which abandons the traditional voting method and uses the weighted probability selection method. Experiments on twenty classic datasets show that the proposed method can effectively mine essential features and enhance classification accuracy compared with original datasets. For instance, the Balance dataset has an average classification accuracy of more than 20% higher than the original dataset on the KNN classifier. The proposed method outperforms alternative feature mapping methods in terms of performance and efficiency on different classifiers in most cases. [ABSTRACT FROM AUTHOR]

[Display omitted] Feature Selection (FS) is considered a crucial procedure for eliminating unnecessary features from datasets. FS is considered a challenging problem that is difficult to solve efficiently due to its combinatorial nature. As the problem size increases, so does the computation time. Recently, researchers have been focusing on metaheuristic FS algorithms. Therefore, this paper introduces an adaptive chaotic dynamic gazelle optimization algorithm (ACD-GOA) with enhanced elite Strategy for FS problems. The ACD-GOA is a novel enhanced version of the recently published GOA, incorporating multiple improved strategies to enhance its search capabilities and convergence speed. The initialization phase and iteration initialization adopt the dynamic opposition learning strategy to avoid premature convergence. Additionally, several enhancement strategies are employed to improve the efficiency of the standard GOA. Adaptive inertia weight and sigmoid function are used to enhance search efficiency. Furthermore, enhanced elite and exchanging information strategies are implemented to maintain population diversity and avoid local solutions, respectively. Experimental evaluations are conducted using various functions, including the twelve CEC2022 standard functions and fourteen FS benchmark datasets. The performance of ACD-GOA is compared with several other metaheuristic algorithms. Statistical tests such as Friedman and Wilcoxon signed-rank tests are used to analyze the experimental data. The experimental results highlight the proposed algorithm's exceptional capacity to overcome problems associated with local minima and expedite the convergence process. The suggested algorithm has been extensively compared to state-of-the-art algorithms, demonstrating significant breakthroughs. The reported accuracy of the proposed algorithm ranges from 0.78 to 1.00 with K-NN classifier. [ABSTRACT FROM AUTHOR]

Class imbalance and concept drifts could deteriorate the performance of classifiers in data stream learning as their co-occurrence presents a complicated learning scenario. This situation becomes more complicated if missing values appear in the data. To tackle this challenge, this paper proposes a novel ensemble method, BINary classification in Incomplete, Non-stationary, and Imbalanced data streams (Bin.INI). As part of Bin.INI, Adapt2Regress builds a linear regression models pool using completely available minority subspaces, monitored under a novel sigmoid-decay function to provide imputation support in a drifting data environment. We use Min++ to balance the data in terms of imbalance complexity, a resampling component proposed earlier. Additionally, the 2-STage Ensemble Pruning technique (2STEP) is proposed which groups the classifiers based upon multiple diversity measures at the first step and selects the best subset using a novel custom metric, Sigmoid-Weighted Imbalance Score, at the second step. By using Adapt2Regress, Min++, and 2STEP, the proposed method can efficiently deal with incomplete, imbalanced, and drifting data in data streams. Experiments are conducted on 350 highly imbalanced data streams containing a variety of concept drifts. 300/350 data streams contain different ratios of missing data (5%–30%) in the minority space. Experiments reveal a noticeable decline in the performance of state-of-the-art approaches when faced with 15% or more missing data, while Bin.INI maintains its performance. [ABSTRACT FROM AUTHOR]

The classification of nonstationary data streams presents a substantial challenge mainly due to the simultaneous presence of class imbalance and concept drift. Ensemble learning is a widely acknowledged approach to address these two problems. Nevertheless, most of the existing methods establish independent strategies for each, ignoring their relationship and interactions, which prevents the corresponding strategies from performing as expected. This paper presents a novel algorithm AB-DES (Adaptive Bagging-based Dynamic Ensemble Selection), which employs bagging technique and dynamic ensemble selection system. In AB-DES, instances are stored to balance subsequent data chunks instead of relying solely on resampling methods. Then, all base classifiers are updated using a new dual adaptive sampling bagging strategy, which can help to reduce the classification inaccuracies involving by class imbalance. In the prediction phase, DES system constructs different ensemble models for different instances. The inclusion of a sliding window-based drift detector makes the above two strategies adapt to concept drift. Experimental results show that AB-DES achieves higher classification accuracy and exhibits enhanced robustness compared to 8 state-of-the-art methods on 10 synthetic datasets and 5 real world datasets with different types of concept drift. • Both of synthesized and stored instances are utilized to achieve data balance. • A new dual adaptive sampling bagging strategy is employed. • The improved DES system dynamically constructs a validation dataset. • An active method based on sliding windows is employed to monitor concept drift. [ABSTRACT FROM AUTHOR]

The skin, one of the most crucial organs of the human body, serves as a barrier between the body and the external environment. Early detection of skin diseases is imperative to reduce mortality rates, as some untreated conditions may progress to skin cancer. Segmentation and classification of lesions represent pivotal and interrelated endeavors within the realm of skin disease diagnosis. For this purpose, this paper presents a comprehensive diagnostic framework for segmentation and classification of skin lesions, which integrates a Legendre multiwavelet transform-based fusion XNet (LW-XNet) with an improved soft attention dense connection convolutional network (ISA-DenseNet). LW-XNet combines the strengths of XNet in fusing different frequency components of images, along with the strong feature representation capability of LW bases with various regularities for overall contextual information and detailed information of dermoscopy images. Furthermore, its encoder devises a LWT channel concatenate (LCC) block to subdivide the image into eight wavelet coefficient feature images and perform concatenated processing on them, enabling it to better differentiate and comprehend the intricate features within dermoscopy images. Finally, ISA-DenseNet is utilized for multi-class classification of the segmented images. Experimental results demonstrate the superiority of the proposed framework over existing segmentation and classification methods. [ABSTRACT FROM AUTHOR]

[Display omitted] Light Detection And Ranging (LiDAR) sensor offers solutions to extract real-time information of vehicle's surroundings and can provide a new opportunity to improve the road safety related issues in mixed traffic conditions. This paper explores fundamental principles and combination of algorithms, to accurately detect, classify and track surrounding vehicles on expressways using a vehicle mounted cost-effective LiDAR sensor in dynamic conditions. The proposed approach employs algorithms such as Density-Based Spatial Clustering of Applications with Noise, Random sample consensus, and Kalman filter on 3D LiDAR data for ground and non-ground points segmentation, object clustering, road boundary identification, vehicle classification and tracking. Trajectory data of the tracked vehicles is used to estimate their relative positions and speeds. The surrounding vehicles were classified into three categories: two-wheelers, four-wheelers and heavy-vehicles. The proposed method achieves an accuracy, precision, recall, and F1-score above 94 %, demonstrating its robustness. The resilience and versatility of the proposed algorithm were validated through comprehensive evaluations on a significant dataset spanning over 8000 km (around 3.6 million frames) of driving data collected under varied environmental scenarios on expressway. Further, various applications of proposed methodology in road safety studies are discussed. The approach can extract real-time micro-level trajectory data for surrounding vehicles on expressways in any lighting condition, which is useful for researchers and can be used in connected and autonomous vehicles. It also provides valuable information for assessing surrogate safety measures and identifying road safety issues and suggesting countermeasures, including advanced and cost-effective driver assistance system applications. [ABSTRACT FROM AUTHOR]

Real-world image classification data usually exhibits a challenging long-tailed distribution, attributed to the inherent difficulty in data collection. Existing ensemble approaches predominantly prioritize the empirical diversification of the ensemble model, sidelining its critical aspects of representation ability. A noticeable gap also exists in theoretical analysis elucidating the intricate relationship between ensemble model diversity and its generalization efficacy. This paper introduces a loosely coupled ensemble framework, Multiple Contrastive Experts (MCE) tailored for long-tailed image classification, aiming to bolster image representation while ensuring diversity within the proposed MCE. Leveraging skill-diverse classification losses, the experts in the ensemble model are able to specialize in different kinds of classes in the long-tailed distribution. An adapted supervised contrastive learning (SCL) loss is introduced to guide training for each feature learning branch to enhance the representation ability of MCE. Through the effective integration of the different losses from all experts, each expert model can be optimized coordinately. Moreover, the relationship between the generalization ability of MCE and its diversity is theoretically revealed against a single classification model. At last, extensive experiments were made over five widely used long-tailed image classification datasets. The results show that our proposed MCE is very competitive against the state-of-the-art methods, while maintaining a relatively lower computational cost. Besides, MCE has also exhibited superior performance with an increasing number of experts. [ABSTRACT FROM AUTHOR]

Multi-task feature selection (MTFS) has been proven effective for reducing the curse of dimensionality in large-scale classification. Many existing MTFS methods assume that all tasks learn concurrently in a static environment without considering the dynamism of tasks in real-world scenarios. However, new tasks emerge dynamically in practical applications, meaning that the aforementioned static assumption is insufficient. In this paper, we construct a dynamic multi-task feature selection framework to achieve feature reduction for constantly arriving new tasks. First, we modify the traditional mapping by shifting from hard labels to soft labels. Unlike the conventional rigid mapping, the new flexible loss function changes the direct mapping strategy to an indirect one. Second, we use the orthogonal regularization term to constrain the independent relationship between new and old tasks. This ensures that the selected relevant features for new tasks differ from prior tasks. Finally, we integrate the flexible loss and the orthogonal regularization term in the dynamic multi-task feature selection framework. Our method outperforms nine other advanced feature selection methods in terms of effectiveness and efficiency across six datasets. For example, the ACC value of our method is almost 1% higher than the next-best method on the large-scale SUN dataset. • A dynamic feature selection is proposed to select features for new tasks. • A flexible loss is presented to incorporate potential knowledge. • An orthogonal constraint is used for selecting task-specific features. [ABSTRACT FROM AUTHOR]

The class imbalance problem is one of the main challenges that hinders classifiers from identifying unknown instances. When class distribution imbalance and class overlap coexist in the same data set, the recognition of minority instances by traditional classifiers will be seriously affected. In the binary classification problem, the relationship between the majority class and the minority class is clear, and the overlap problem between classes can be well solved by removing the majority instances in the class-overlap area. However, for multi-class imbalance problems, the relationships between classes will become complex. Clustering techniques are effective methods for identifying overlapping instances; however, a single clustering method cannot effectively identify clusters of different shapes. Therefore, in this paper, a H eterogeneous C lustering E nsemble learning method for M ultiple C lass-overlap D etection (HCE-MCD) is proposed for multi-class imbalance problems. This method uses a genetic algorithm to select a combination of heterogeneous clustering techniques with the best fitness function from a pool of clustering algorithms and uses majority voting to fuse the results of the clustering algorithm. Furthermore, instances located near the decision boundary are more likely to be overlapping instances. Therefore, according to the clustering results of the heterogeneous clustering ensemble method, we can identify the overlapping instances existing in the data set, thereby eliminating the class overlap problem in the multi-class dataset. Experimental results on 19 open-access datasets show that our proposed method outperforms or partially outperforms state-of-the-art research protocols. [ABSTRACT FROM AUTHOR]

Feature selection is a fundamental step in machine learning, serving to reduce dataset redundancy, accelerate training speed, and improve model quality. This is particularly crucial in high-dimensional datasets, where the excess of features presents challenges for pattern recognition and data analysis. Recent methods proposed for high-dimensional data are often tailored for specific domains, leaving a lack of consensus on a universally recommended solution for general use cases. This paper proposes a hybrid feature selection approach using a multi-objective genetic algorithm to enhance classification performance and reduce dimensionality across diverse classification tasks. The proposed approach narrows the search space of possible relevant features by exploring the combined outputs of classical feature selection methods through novel genetic algorithm operators. This enables the evolution of combined solutions potentially not explored by the original methods, generating optimized feature sets in a process that adapts to different data conditions. Experimental results demonstrate the effectiveness of the proposed method in high-dimensional use cases, offering improved classification performance with reduced feature sets. In summary, our hybrid method offers a promising solution for addressing the challenges of high-dimensional datasets by enhancing classification performance in varying domains and data conditions. • Hybrid method optimizes feature selection with multi-objective genetic algorithm. • Achieves improved classification metrics with optimized and smaller feature sets. • Combines multiple feature selection mechanisms for superior results. • Efficiently represents information in complex datasets from multiple domains. [ABSTRACT FROM AUTHOR]

• 140 papers that deal with the blocking flow shop scheduling problem are reviewed. • Papers are classified according to the purpose and objective function adopted. • Tables allow the reader to find the interested literature. • Based on the gaps discovered, suggestions for future research are pointed out. This paper presents a literature review on the m -machine flow shop scheduling problem with blocking conditions. In total, 139 papers are reviewed and classified, ranging from 1969 up to early 2019. Results showed that makespan was the objective function most adopted by the researchers, with 62% of the total covered, followed by multi-objective based (12%), total flow time-based (11%), due date-based (7%), stochastic based functions (6%) and cycle time (3%). Regarding the purpose of the paper, approximately 92% of the papers proposed solution methods, where 76% of the papers developed heuristic methods and 16% of exact methods, and 8% of the papers considered the analysis of the problem and literature reviews. Directions for future researches include the proposition of solution methods for mono-objective functions as total flow time-based and due date-based, development of solution methods for the m -machine flow shop with RCb and RCb* constraints and adoption of more than one additional constraint to the problem. [ABSTRACT FROM AUTHOR]

• This paper proposes the Recursive General Regression Neural Network Oracle. • The algorithm was compared to several other classifiers. • It had the highest performance metrics such as accuracy and area under the curve. • The algorithm had also bested the literature in the performance metrics. This research introduces the Recursive General Regression Neural Network Oracle (R-GRNN Oracle) and is demonstrated on several binary classification datasets. The traditional GRNN Oracle classifier (Masters et al., 1998) combines the predictive powers of several machine learning classifiers by weighing the amount of error each classifier has on the final predictions. Each classifier is assigned a weight based on the percentage of errors it contributes to the final predictions as the classifiers evaluate the dataset. The proposed R-GRNN Oracle is an enhancement to the GRNN Oracle in which the proposed algorithm consists of an oracle within an oracle – where the inner oracle acts as a classifier with its own predictions and error contribution. By combining the inner oracle with other classifiers, the R-GRNN Oracle produces superior results. The classifiers considered in this study are: Support Vector Machine (SVM), Multilayer Perceptron (MLP), Probabilistic Neural Network (PNN), Gaussian Naïve Bayes (GNB), K-Nearest Neighbor (KNN), and Random Forest (RF). To demonstrate the effectiveness of the proposed approach, several datasets were used, with the primary one being the publicly available Spambase dataset. The predictions of SVM, MLP, KNN, and RF were used to create the first GRNN Oracle, which was then enhanced with the high performances of SVM and RF to create the second oracle, the R-GRNN Oracle. The combined recursive model was 93.24% accurate using 10-fold cross validation, higher than the 91.94% of the inner GRNN Oracle and the 91.29% achieved by RF, the highest performance by a stand-alone classifier. The R-GRNN Oracle was not only the most accurate, but it also had the highest AUC, sensitivity, specificity, precision, and F1-score (97.99%, 91.86%, 94.40%, 93.28%, and 92.57%, respectively). The research contribution of this paper is introducing the concept of recursion (a concept not fully explored in machine learning models and applications) and testing this structure's ability on further enhancing the performance of the traditional oracle. The recursive model has also been applied to several other datasets: The Human Resources, Bank Marketing, and Monoclonal Gammopathy of Undetermined Significance (MGUS) datasets. The results of these implementations are summarized in this paper. [ABSTRACT FROM AUTHOR]

• Empirical demonstration of Sparse classification methods on text data. • Minimum residual error and class wise dictionaries based methods are most efficient. • Proposed a dynamic dictionary refinement algorithm. • Proposed work improves computational speed and classification performance. • Dictionary refinement procedure addressed the problem of over-reliance on training data. Classification of text based on its substance is an essential part of analysis to organize enormously large text data and to mine the salient information contained in it. It is gaining greater attention with the surge in the volume of on-line data available. Classical algorithms like k -NN (k-nearest neighbor), SVM (Support Vector Machine) and their variations have been observed to yield only reasonable results in addressing the problem, leaving enough room for further improvement. A class of algorithms commonly referred to as Sparse Methods has been emerged recently from compressive sensing and found numerous effective applications in many areas of data analysis and image processing. Sparse Methods as a tool for text analysis is an alley that is largely unexplored rigorously. This paper presents exploration of sparse representation-based methods for text classification. Based on the success of sparse representation based methods in different areas of data analysis, we intuitively hypothesized that it should work well on text classification problems as well. This paper empirically reinforces the hypothesis by testing the method on Reuters and WebKB data sets. The empirical results on Reuters and WebKB benchmark data show that it can outperform classical classification algorithms like SVM and k -NN. It has been observed that obtaining the basis of representation and sparse codes are computationally costly operations affecting the performance of the system. We also propose a class-wise dictionary refinement algorithm and dynamic dictionary selection algorithm to make sparse coding faster. The addition of dictionary refinement to the classification system not only reduces the time taken for sparse coding but also gives improved classification accuracy. The outcomes of the study are empirical verification of sparse representation classifier as a text classification tool and a computationally efficient solution for the bottleneck operation of sparse coding. [ABSTRACT FROM AUTHOR]

• The prediction task of CoIL challenge 2000 is addressed in the paper. • Three different methods are proposed for direct mailing problem of CoIL challenge 2000. • The proposed methods outperform the method proposed by the winner of the challenge. • The proposed methods overcome, also, the unbalanced dataset issue of the problem. This paper presents some approaches based on data mining techniques to solve the prediction task of Computational Intelligence and Learning (CoIL) Challenge 2000. The prediction task of the contest is a direct mailing problem and the goal is to improve its response rate. The main issue in this competition is the incompatibility of the dataset in which the distribution of the classes of the target attribute is highly unbalanced. This in turn causes high error rate in identifying the minority class samples. Three different level methods including data-level, algorithm-level, and hybrid method are used to overcome this issue. The specificity, sensitivity, precision-recall, and ROC criteria are employed to compare the performance of the methods. Among the methods proposed in this paper, the best one performs much better than the winner of the competition. [ABSTRACT FROM AUTHOR]

Highlights • A novel modeling strategy for Decision Trees is proposed. • Interactive basis functions induce oblique and non-linear decision boundaries. • Benchmarking shows potential gains in accuracy and/or parsimony of models. • Empirical examples illustrate the application of the modeling strategy. • Decision charts are introduced to improve the interpretability of models. Abstract Decision Trees (DTs) are a machine learning technique widely used for regression and classification purposes. Conventionally, the decision boundaries of Decision Trees are orthogonal to the features under consideration. A well-known limitation of this is that the algorithm may fail to find optimal partitions, or in some cases any partitions at all, depending on the underlying distribution of the data. To remedy this limitation, several modifications have been proposed that allow for oblique decision boundaries. The objective of this paper is to propose a new strategy for generating flexible decision boundaries by means of interactive basis functions (IBFs). We show how oblique decision boundaries can be obtained as a particular case of IBFs, and in addition how non-linear decision boundaries can be induced. One attractive aspect of the strategy proposed in this paper is that training Decision Trees with IBFs does not require custom software, since the functions can be precalculated for use in any existing implementation of the algorithm. Since the underlying mechanisms remain unchanged there is no substantial computational overhead compared to conventional trees. Furthermore, this also means that IBFs can be used in any extensions of the Decision Tree algorithm, such as evolutionary trees, boosting, and bagging. We conduct a benchmarking exercise to understand under which conditions the use of IBFs can improve model the performance. In addition, we present three empirical applications that illustrate the approach in classification and regression. As part of discussing the empirical applications, we introduce a device called decision charts to facilitate the interpretation of DTs with IBFs. Finally, we conclude the paper by outlining some directions for future research. [ABSTRACT FROM AUTHOR]

Highlights • Novel Neutrosophic Rule-based Classification System (NRCS) models are proposed. • Our dataset has 553 drugs that were bio-transformed into liver. • Three rough set-based algorithms are used for features selection. • Three sampling algorithms are used to obtain balanced data. • The proposed model obtained promising results. Abstract Measuring toxicity is an important step in drug development. However, the current experimental methods which are used to estimate the drug toxicity are expensive and need high computational efforts. Therefore, these methods are not suitable for large-scale evaluation of drug toxicity. As a consequence, there is a high demand to implement computational models that can predict drug toxicity risks. In this paper, we used a dataset that consists of 553 drugs that biotransformed in the liver. In this data, there are four toxic effects, namely, mutagenic, tumorigenic, irritant and reproductive effects. Each drug is represented by 31 chemical descriptors. This paper proposes two models for predicting drug toxicity risks. The proposed models consist of three phases. In the first phase, the most discriminative features are selected using rough set-based methods to reduce the classification time and improve the classification performance. In the second phase, three different sampling algorithms, namely, Random Under-Sampling, Random Over-Sampling, and Synthetic Minority Oversampling Technique (SMOTE) are used to obtain balanced data. In the third phase, the first proposed model employs the Neutrosophic Rule-based Classification System (NRCS), and the second model uses Genetic NRCS (GNRCS) to classify an unknown drug into toxic or non-toxic. The experimental results proved that the proposed models obtained high sensitivity (89–93%), specificity (91–97%), and GM (90–94%) for all toxic effects. Overall, the results of the proposed models indicate that it could be utilized for the prediction of drug toxicity in the early stages of drug development. [ABSTRACT FROM AUTHOR]

Highlights • Mammographic masses classification is dominated by ANN, SVM and KNN. • Grammars and syntactic approaches have not been used so far. • Genetic algorithm was the most used technique to reduce feature dimensionality. • Most of the published papers are database-dependent. • Results are not comparable due to the lack of a common benchmark. Abstract Breast cancer is one of the most common and deadliest cancers that affect mainly women worldwide, and mammography examination is one of the main tools to help early detection. Several papers have been published in the last decades reporting on techniques to automatically recognize breast cancer by analyzing mammograms. These techniques were used to create computer systems to help physicians and radiologists obtain a more precise diagnosis. The objective of this paper is to present an overview regarding the use of machine learning and pattern recognition techniques to discriminate masses in digitized mammograms. The main differences we found in the literature between the present paper and the other reviews are: 1) we used a systematic review method to create this survey; 2) we focused on mass classification problems; 3) the broad scope and spectrum used to investigate this theme, as 129 papers were analyzed to find out whether mass classification in mammograms is a problem solved. In order to achieve this objective, we performed a systematic review process to analyze papers found in the most important digital libraries in the area. We noticed that the three most common techniques used to classify mammographic masses are artificial neural network, support vector machine and k-nearest neighbors. Furthermore, we noticed that mass shape and texture are the most used features in classification, although some papers presented the usage of features provided by specialists, such as BI-RADS descriptors. Moreover, several feature selection techniques were used to reduce the complexity of the classifiers or to increase their accuracies. Additionally, the survey conducted points out some still unexplored research opportunities in this area, for example, we identified that some techniques such as random forest and logistic regression are little explored, while others, such as grammars or syntactic approaches, are not being used to perform this task. [ABSTRACT FROM AUTHOR]

This paper is dedicated to solving the problem of Advanced Persistent Threat (APT) attack and defense in the Industrial Internet of Things (IIoT). Due to the diversity of IIoT equipment and the inconsistency of protection capabilities, it is difficult for the existing uniform defense strategy and the random defense strategy to achieve ideal results. Considering that both attackers and defenders aim to achieve maximum benefits by paying the minimum cost, as well as the differences between devices, this paper proposes an equipment classification based differential game method for APT in IIoT. Firstly, all equipment is divided into two categories according to their protective capabilities. Secondly, the APT attack and defense process is mathematically described, and the corresponding differential game problem is formulated and analyzed theoretically. Finally, the theoretical results of this method are verified by various experiments, including the comparisons with the uniform defense strategy, the random defense strategy, and the latest model. • A new equipment-based evolutionary dynamics model is proposed. • A necessity system is established to narrow the scope of searching for Nash equilibrium. • An algorithm is designed to generate the Nash equilibrium. [ABSTRACT FROM AUTHOR]

This paper provides the first review to date which gathers, describes, and assesses, to the best of our knowledge, all available publications on automating cerebral microbleed (CMB) detection. It provides insights into the current state of the art and highlights the challenges and opportunities in this topic. By incorporating the best practices identified in this review, we established guidelines for the development of CMB detection systems. We are confident that these guidelines can serve as a foundation for further research. CMB detection is a crucial but challenging task that can be laborious for radiologists. With the increasing popularity of magnetic resonance imaging (MRI), the ability to detect CMBs has improved, but there is still a need to automate this process to enhance its efficiency and accuracy. A high prevalence of CMBs is closely associated with cognitive dysfunction, diabetes, hypertension, an increased risk of stroke, and intracerebral hemorrhage. It is alarming to note that strokes, Alzheimer's disease, and Diabetes mellitus have secured their position as the second, seventh, and ninth most common causes of death worldwide, respectively. Moreover, CMBs are sometimes found in association with other pathologies and indicate a range of pathological processes in the cerebral vessels. Thus, it is essential to enhance the quality of diagnostics to facilitate prompt identification and treatment of these potentially life-threatening conditions. In this paper, we aimed to systematize the existing knowledge and best practices in automatic CMB detection, from fundamental information about CMBs and MRI image data, through employed datasets and CMB detection and verification algorithms, to methods of result evaluation. This can serve as a starting point for future research and the development of a CMB detection system that is practically applicable in medicine, leading to enhanced patient treatment outcomes. • There is a need for automatization of laborious CMB detection process. • Automatic CMB detection suffers from high production of false-positive predictions. • The scarcity of publicly available CMB datasets hinders research comparison. [ABSTRACT FROM AUTHOR]

Accurate grading of gliomas and classification of various brain tumor types are essential for diagnosing and treating brain cancer. In this work, a new combined network structure has been designed and implemented to effectively capture detailed and global feature information for precise pattern recognition and classification. It utilizes the advantages of both transformers and CNNs. The structure is constructed using five blocks that have been carefully designed and properly located to achieve high-accuracy classifications. Among these blocks, there exist not only blocks designed homogeneously, including only CNNs or transformers, but also heterogeneously designed blocks, including an effective combination of both CNNs and transformers. Also, for fair comparative evaluations, state-of-the-art brain tumor classifications have been applied to the same datasets, and their capabilities have been evaluated using the same assessment criteria. Therefore, the main contributions of this paper are fourfold: (i) Introducing a novel network utilizing the advantages of both transformers and CNN structures. (ii) Demonstrating the effectiveness of the network in pattern recognition for glioma grading and multi-class classifications of brain tumors. (iii) Providing fair comparative assessments of recent methods, trained and tested with the same datasets and evaluated using the same metrics. (iv) Presenting experimental results demonstrating that the proposed approach achieves higher accuracy in both glioma grading (99.21%) and brain tumor classification (98.66%) compared to recent methods. [ABSTRACT FROM AUTHOR]

In the last decade, deep learning research has made amazing progress, and the performance of deep neural networks trained on large-scale datasets is excellent, and in some respects even exceeds that of humans. However, when deep neural networks are faced with long-tailed datasets, their performance deteriorates dramatically. Most of the existing long-tailed classification methods focus only on the classification, leading to the model being biased toward majority classes and resulting in poor performance on minority classes. Calibration, known for enhancing model reliability, can help mitigate this problem in the long-tailed setting by adjusting class probabilities to align with the true distribution, ensuring better consideration of minority classes. In this paper, a novel dynamic prior probability network (DPPN) is proposed, which enhances classification by incorporating unbiased and calibrated confidence estimates. We experiment on four widely used long-tailed datasets and compared the proposed method DPPN with eight methods, including a benchmark method and seven state-of-the-art methods. The experimental results show that DPPN achieved significant improvements in confidence estimation and classification performance on all four benchmark datasets. Our code is available at https://github.com/zxsong999/DPPN. • A new dynamic prior probability network DPPN is proposed. • DPPN can pay more attention to the tail category. • DPPN significantly improves the recognition accuracy of tail categories. [ABSTRACT FROM AUTHOR]

Tagging knowledge concepts for course problems is essential for intelligent tutoring systems. Traditional manual tagging methods, usually performed by domain experts, are time-consuming and subject to individual biases. Consequently, research on automatic tagging technology is of substantial practical importance. Recently, text classification techniques have been applied to this task; however, these methods are inadequate for math problems due to their complexity, which includes formulaic content and hierarchical relationships among knowledge concepts. Although large language models (LLMs) have also been explored for this purpose, their generative nature and high computational cost pose challenges for direct application in tutoring systems. In this paper, we propose an automatic knowledge concept tagging model LHABS based on RoBERTa. This model integrates hierarchical label-semantic attention, which captures hierarchical knowledge concepts information, and multi-label smoothing, which combines textual features to help reduce overfitting, thus enhancing text classification performance. Our experimental evaluation on four datasets demonstrates that our model outperforms state-of-the-art methods. We also validate the effectiveness of hierarchical label-semantic attention and multi-label smoothing through our experiments. The code and data are available at: https://github.com/xuqiang124/atmk_system. [Display omitted] • A RoBERTa-based automatic tagging model LHABS is proposed to handle Chinese math problems. • Hierarchical label-semantic attention and multi-label smoothing mechanisms are incorporated to enhance performance. • A real-world Chinese dataset of 22,498 math problems is constructed. • LHABS outperforms state-of-the-art models on four datasets. [ABSTRACT FROM AUTHOR]

Big data comes with the challenge of containing irrelevant and redundant information (i.e., features). Given that a single objective cannot fully capture a feature's relevance, a Many-Objective Feature Selection (MOFS) approach able to accommodate various relevant perspectives is preferred for identifying the most appropriate features in a given context. However, MOFS produces a large set of solutions whose interpretability has been largely overlooked. First, we demonstrate the relevance of MOFS and establish its necessity by considering up to six objectives using a genetic algorithm and Naive Bayes on ten datasets for classification tasks. Then, we propose a novel methodology to improve the interpretability of MOFS results in order to support the data scientist in selecting the subset of features pertinent to their use case. Our methodology is instantiated as an intuitive and interactive dashboard that provides insights into the results beyond the pure numerical representation of the objectives being considered and evaluated with 50 participants. The outcome shows that it addresses the need for a methodological approach and comprehensive visualization to achieve interoperability. • This paper presents an empirical analysis of Many-Objective Feature Selection (MOFS). • A novel methodology for interpreting MOFS results is proposed. • The methodology is implemented via an interactive dashboard. • Statistical comparison of the methodology versus the tabular method with 50 subjects. • Sensitivity analysis of the methodology to changes in objective weights. [ABSTRACT FROM AUTHOR]

Out-of-distribution detection (OOD detection) aims to detect test samples drawn from a distribution that is different from the training distribution, in order to prevent models trained on in-distribution (ID) data from providing unavailable outputs. Current OOD detection systems typically refer to a single-domain class-balanced assumption that both the training and testing sets belong to the same domain and each class has the same size. Unfortunately, most real-world datasets contain multiple domains and class-imbalanced distributions, which severely limits the applicability of existing works. Previous OOD detection systems only focus on the semantic gap between ID and OOD samples. Besides the semantic gap, we are faced with two additional gaps: the domain gap between source and target domains, and the class-imbalance gap between different classes. In fact, similar objects from different domains should belong to the same class. In this paper, we introduce a realistic yet challenging setting: class-imbalanced cross-domain OOD detection (CCOD), which contains a well-labeled (but usually small) source set for training and conducts OOD detection on an unlabeled (but usually larger) target set for testing. We do not assume that the target domain contains only OOD classes or that it is class-balanced: the distribution among classes of the target dataset need not be the same as the source dataset. To tackle this challenging setting with an OOD detection system, we propose a novel uncertainty-aware adaptive semantic alignment (UASA) network based on a prototype-based alignment strategy. Specifically, we first build label-driven prototypes in the source domain and utilize these prototypes for target classification to close the domain gap. Rather than utilizing fixed thresholds for OOD detection, we generate adaptive sample-wise thresholds to handle the semantic gap. Finally, we conduct uncertainty-aware clustering to group semantically similar target samples to relieve the class-imbalance gap. Extensive experiments on three challenging benchmarks (Office-Home, VisDA-C and DomainNet) demonstrate that our proposed UASA outperforms state-of-the-art methods by a large margin. • We pose a practical yet challenging setting for an OOD detection system. • We propose a novel UASA network with four modules to handle three gaps. • Extensive experiments on three challenging benchmarks show the effectiveness of UASA. [ABSTRACT FROM AUTHOR]

Surface defect detection plays a vital role in modern industry product quality control. However, due to the large variation in the shape and scale of defects, achieving accurate detection is a great challenge. Existing CNN-based detection methods perform well in local modeling, but it is difficult to extract accurate features when dealing with defects with significant changes in shape and scale. To overcome these challenges, this paper leverages the long-range dependency modeling advantages of the Transformer architecture and proposes an innovative end-to-end detection network named FC-DETR. Firstly, a foreground supervision module (FSM) is introduced to enhance the focus on foreground defect features and reduce the computational of the encoder. Secondly, a cascade refined hybrid matching (CRHM) strategy is proposed to increase the number of positive samples during training and avoid NMS post-processing in predicting. Thirdly, the IA-BCELoss is introduced as the classification loss function, coupling classification scores with the IoU of predicted and ground truth boxes to obtain high-quality detection boxes. Finally, the effectiveness and advancement of the proposed method are validated on three public defect detection datasets: NEU-DET, GC10-DET, and PCB defect dataset. The experimental results show that the mAP50 of the proposed method on NEU-DET, GC10-DET, and PCB datasets is 83.7%, 83.6%, and 98.7%, respectively. Compared with the CNN-based detection method, it is significantly improved. Ablation experiments further verify the effectiveness of the proposed modules. [ABSTRACT FROM AUTHOR]

Domain adaptation methods using electroencephalography (EEG) play an important role in cross-subject emotion recognition. However, enhancing the generalizability of such models remains challenging. This paper proposes a domain adversarial neural network using multiple adversarial tasks (DANN-MAT). Multiple emotion-unrelated classification tasks are designed to adversarially challenge an emotion classifier, thereby preserving emotion-related features and eliminating irrelevant information. The results demonstrate that employing multiple adversarial tasks improves the generalizability of the developed model, and utilizing more adversarial tasks leads to higher accuracy in cross-subject emotion classification. Our method achieves state-of-the-art results on the SEED and SEED-IV datasets, and the superior stability of the algorithm is proven via parameter comparison experiments. • A domain adversarial neural network using multiple adversarial tasks is proposed. • Multiple adversarial tasks could improve the generation of emotion classifier. • The method improves the performance in cross-subject EEG emotion recognition. [ABSTRACT FROM AUTHOR]

In real-world scenarios where resources for evaluating expensive optimization problems are limited and the reliability of trained models is hard to assess, the quality of the non-dominated front formed by algorithms tends to be low. This paper proposes a metric-based surrogate-assisted evolutionary algorithm for multi-objective expensive optimization, incorporating a novel model management strategy that integrates a regeneration mechanism. This approach aims to achieve a well-balanced convergence and diversity, facilitating the attainment of high-quality non-dominated fronts to address expensive multi-objective optimization problems. The model management strategy, based on metrics, comprehensively evaluates the reliability of the classification model and selects appropriate strategies for offspring selection. Moreover, through significance analysis of the population, the regeneration mechanism identifies high-quality dimensions for regenerating offspring. The algorithm maximizes the utilization of the classification model to guide the generation and selection of offspring in the population. Experiments on DTLZ, MaF, WFG, and the high-dimensional portfolio optimization problem demonstrate that the proposed algorithm outperforms nine state-of-the-art surrogate-assisted evolutionary algorithms, highlighting its superior performance across various scenarios. [Display omitted] • Propose a surrogate-assisted algorithm for expensive multi-objective optimization. • Apply dynamic offspring selection based on metrics in the optimization process. • Introduce an offspring regeneration mechanism for better model and population use. • Enhance diversity with linear weight-assisted generic front modeling. • Achieve better non-dominated fronts on DTLZ, MaF, and WFG open datasets. [ABSTRACT FROM AUTHOR]

In many image classification scenarios where knowledge distillation (KD) is applied, multiple users need to train various student models that conform to the device's computational limitations at different times. However, due to security concerns, existing data-free KD (DFKD) methods discard the trained generator after completing a task, preventing its reuse in training subsequent students and resulting in wasted resources. Furthermore, the unlabeled nature of fake images eliminates the hard loss that plays a crucial role in data-driven KD. This paper introduces the reusable generator DFKD with hard loss simulation (RG-HLS) to address these challenges. RG-HLS employs a generator consisting of two generative networks and a decoder. The decoder and one of the generative networks can be reused in subsequent tasks. The adversarial example attack is also introduced to mimic the behavior of the teacher model in predicting real images, ensuring that DFKD exhibits a hard loss similar to that of data-driven KD. ,Extensive experiments across CIFAR-10, CIFAR 100, MNIST, SVHN, and Tiny-ImageNet datasets validate the proposed method's superior performance. The reusable generator accelerates convergence without compromising security, and integrating the hard loss further boosts accuracy. Code is available at https://github.com/sunyafeng-jlu/RG-HLS. • Designed a reusable generator for data-free knowledge distillation (DFKD). • Simulated teacher's behavior with adversarial example attacks to obtain hard loss. • Proposed a novel DFKD method for image classification. • The reusable generator speeds up convergence without breaking the security of DFKD. • The hard loss simulation improves the classification accuracy of DFKD. [ABSTRACT FROM AUTHOR]

Few-Shot Object Detection (FSOD) is a significant application of few-shot learning in object detection tasks. Its primary objective is to enable the model to quickly acquire the ability to detect novel categories through a limited number of annotated samples. However, such a small number of training samples is insufficient for the model to be fully trained, which leads to difficulties in completely decoupling the regression and classification tasks, and confusion among categories. To address this issue, this paper proposes to remap features into location and class spaces. By imposing task orthogonality and class orthogonality constraints on the fine-tuning process, task decoupling and class decoupling are achieved. Furthermore, we design and implement a progressive fine-tuning strategy that iteratively optimizes in a lagged and progressive manner through the combination of training samples and historical features, mitigating the risks of catastrophic forgetting and overfitting. Experimental results demonstrate that our method effectively improves the model's performance in FSOD tasks, achieving state-of-the-art metrics on the PASCAL VOC and MS COCO benchmarks. • Task orthogonality to decouple regression and classification. • Class orthogonality to enhance inter-class discrimination. • Progressive fine-tuning strategy to mitigate forgetting and overfitting. • Implement orthogonal progressive network to achieve SOTA results. [ABSTRACT FROM AUTHOR]

Graph convolutional network has emerged as a focal point in machine learning because of its robust graph processing capability. Most existing graph convolutional network-based approaches are designed for single-view data, yet in many practical scenarios, data is represented through multiple views. Moreover, due to the complexity of multiple views, normal graph generation methods cannot mitigate redundancy to generate a high quality graph. Although the ability of graph convolutional network is undeniable, the quality of graph directly affects its performance. To tackle the aforementioned challenges, this paper proposes a multi-scale graph generation deep learning framework, called multi-scale semi-supervised graph generation based multi-view classification, consisting of two modules: edge sampling and path sampling. The former aims to generate an adjacency graph by selecting edges based on the maximum likelihood among graphs from different views. Meanwhile, the latter seeks to construct an adjacency graph according to the characteristics of paths within the graphs. Finally, the statistical technique is employed to extract commonality and generate a fused graph. Extensive experimental results robustly demonstrate the superior performance of our proposed framework, compared to other state-of-the-art multi-view semi-supervised approaches. • Two alternative sampling methods are provided to collect the structural information, and then a statistical technique is employed to generate a fused graph. • An efficient graph generation framework has been developed to construct graphs in multi-view learning. • Comprehensive evaluations of six real-world datasets show the superiority of the proposed framework. [ABSTRACT FROM AUTHOR]

Domain adaptation (DA) aims to transfer knowledge of the labeled source domain to the unlabeled target domain. Current DA methods learn domain-invariant features by aligning source and target feature spaces by distribution discrepancy minimization or adversarial training. However, existing DA methods fail to consider the semantic-level alignment, which can lead to the distortion of semantic feature structures for the target tasks. In this paper, we propose a novel domain adaptation method, called Deep Joint Subdomain Alignment (DJSA), which jointly conducts intra-subdomain and inter-subdomain alignment for fine-grained classification of target tasks. DJSA obtains the pseudo labels of unlabeled target samples by measuring the similarity between the target samples and source category centers, and then DJSA calculates an intra-subdomain loss to reduce the difference within the same class and increase the difference between different classes in both the source domain and the target domain, and DJSA utilizes a Local Maximum Mean Discrepancy (LMMD) metric to measure the inter-subdomain discrepancy between the source and target domain, and minimizes the discrepancy to align different domains. Furthermore, extensive experimental results demonstrate the effectiveness and superiority of our proposed method in unsupervised domain adaptation. • A new pseudo label acquisition method for subdomain adaptation is proposed. • A optimization method for joint intra- and inter-subdomain adaptation. • Experimental results on public dataset show the effectiveness of our method. [ABSTRACT FROM AUTHOR]

Mobile crowdsensing is a recent model in which a group of mobile users uses their smart devices such as smartphones or wearable devices to cooperatively perform a large-scale sensing task. In this paper, a novel model will be introduced for recognizing/classifying human activities that were collected from sensor units on the chest, legs, and arms. The proposed model employed the k -Nearest Neighbor ( k -NN) classifier which is one of the most common classifiers. k -NN has only one parameter, k , to determine the number of selected nearest neighbors to the test or unknown samples for predicting the class labels of the unknown samples. Searching for the value of k which has a great impact on the classification performance is difficult especially with high dimensional data. This paper employs the Particle Swarm Optimization (PSO) algorithm to search for the optimal value of the k parameter in the k -NN Classifier. This paper shows first experimentally how the PSO in the proposed algorithm searches for the optimal value of k parameter to reduce the misclassification rate of the k -NN classifier. Then, in the second experiment, ten standard datasets are utilized to benchmark the performance of the proposed algorithm. For verification, the results of the PSO- k NN algorithm are compared with two well-known algorithms: Genetic Algorithm (GA) and Ant Bee Colony Optimization (ABCO). In the third experiment, the proposed PSO- k NN algorithm was employed for recognizing human activities. The experimental results proved that the PSO- k NN algorithm is able to find the optimal or near optimal value(s) of the k parameter which enhances the accuracy of k -NN classifier. The results also demonstrated lower error rates compared when GA and ABCO algorithms. [ABSTRACT FROM AUTHOR]