Search

As a new technology for model compression, self-knowledge distillation (SKD) avoids the large computational overhead of training the teacher model seen with traditional knowledge distillation. However, existing SKD methods pay attention to the knowledge transfer between deep and shallow layers of the network but ignore the mutual learning between shallow branches. This paper proposes a many-objective evolutionary self-knowledge distillation framework (MaOESKD) to guide the knowledge fusion between branches in the SKD neural network. This framework embeds an optimization module and a temporary branch into the multi-branch SKD network. The optimization module includes a many-objective adaptive weight optimization model (MaAWOM) and an many evolutionary optimization algorithm based on multi-strategy consensus mechanism (MaOEA-MCM); meanwhile, the temporary branch performs linear weighted fusion. In the MOBWOM, the weight of different branches in knowledge fusion is taken as the decision variable, and the mutual information, covariance, KL divergence between branch output features, and the total information of each branch are taken as the optimization objective. The MSCMEA integrates several state-of-the-art individual selection strategies in the field of evolutionary algorithms. It includes shift density estimation (SDE), penalized boundary intersection (PBI), balanced fitness estimation (BFE), and adaptive position transformation (APT). The accuracy of MOESKD achieves 99.70, 95.74 and 78.21 in MNIST, CIFAR-10 and CIFAR-100. [ABSTRACT FROM AUTHOR]

In this paper, we introduce a decentralized Event-Triggered fault-tolerant echo-state network (ESN) direct adaptive controller for uncertain interconnected nonlinear systems in pure-feedback form. The proposed controller addresses input saturation, actuator faults, external disturbances, and unavailable states for measurement. Unlike the existing works in the literature that adapts the ESN weights using the tracking error, our method employs the control error that is estimated using a fuzzy inference system, to derive the adaptation laws. The complexity explosion typically seen with recursive back-stepping and Dynamic Surface Control (DSC) designs is completely avoided. Our control scheme addresses four types of state-dependent actuator faults: bias, drift, loss of accuracy, as well as loss of effectiveness. The ESNs approximate unknown ideal control laws, while robust terms are added to enhance the stability of the closed-loop system. Stability analysis ensures that tracking errors converge, in finite time, to a small compact set near the origin due to the strictly positive real (SPR) property. Simulation results of a quadrotor UAV and a mathematical system are presented. A comparative analysis is performed with another approach to emphasize the effectiveness of the proposed method. • An ESN adaptive controller is proposed for interconnected pure-feedback unknown nonlinear systems with finite-time stability. • The input saturation, actuators faults, triggering events and complexity explosion are dealt with using state transformation. • The adaptive laws of the ESN are derived for the first time using control-errors instead of tracking-errors. • The parameters of the adaptive laws are highly reduced due to the use of only one ESN regardless of system's order. [ABSTRACT FROM AUTHOR]

Graph neural networks (GNNs) have demonstrated remarkable performance in diverse graph-related tasks, including node classification, graph classification, link prediction, etc. Previous research has indicated that GNNs are vulnerable to membership inference attacks (MIA). These attacks enable malevolent parties to deduce whether the data points are part of the training set by identifying the output distribution, giving rise to noteworthy privacy apprehensions, especially when the graph contains sensitive data. There have been some studies to defend against graph MIA so far, but they have issues like high computational cost and decreased model accuracy. In this paper, we introduce a novel defense framework called MaskArmor , designed to bolster the privacy and security of GNNs against MIA. The MaskArmor framework encompasses four distinct masking strategies: AdjMask, DTMask, ATMask, and SigMask. These strategies leverage message-passing mechanisms, distillation temperature, hybrid masking, and the Sigmoid function, respectively. The MaskArmor framework effectively obscures the distribution of the model on both the training and non-training samples, rendering it challenging for attackers to ascertain whether particular samples have undergone training. Additionally, MaskArmor sustains the model's precision with negligible computational overhead. Our experiments are implemented across seven benchmark datasets and four GNN networks against shadow-based and threshold-based MIAs, showcasing that MaskArmor substantially heightens GNNs' resilience against MIA while simultaneously preserving accuracy on the initial tasks. It also demonstrates adeptness in countering threshold-based MIA through strategies like AdjMask and ATMask. Exhaustive experimental results substantiate that MaskArmor outperforms alternative existing approaches, maintaining effectiveness and applicability across diverse datasets and attack scenarios. [ABSTRACT FROM AUTHOR]

This paper introduces a novel machine learning approach, called BBATDD (Boosting-Based Algorithm Trained with Drift Detector), that departs from traditional epoch-based methods by adopting a data stream generation strategy from the original dataset. The proposed method is versatile, applicable to a broad range of deep structures and other machine learning architectures. The key to its effectiveness is the conversion of original data into a streaming sequence, which is then fed into the neural network structure. To monitor the learning process and prevent overfitting, a drift detector, commonly used in stream data mining, is employed. Additionally, two methods, ELB and ENT, are introduced to prioritize problematic data elements during the learning process, enhancing overall model performance. Extensive simulations on benchmark datasets (MNIST and CIFAR) validate that BBATDD achieves higher accuracy with significantly fewer batches, without sacrificing generalization capability. For instance, when applied to the MNIST dataset and processing 10 000 batches while comparing loss function values, our approach demonstrates a remarkable 45.6% improvement. [ABSTRACT FROM AUTHOR]

Unsupervised Knowledge Amalgamation (UKA) trains a versatile student model with an unlabeled dataset to handle joint objectives of multiple off-the-shelf teacher models, each specialized for distinct tasks. Existing research primarily focuses on facilitating the learning of discriminative knowledge by student models from their teacher models within a latent feature space, but often overlooks the transferability of such knowledge to downstream tasks. In this paper, we propose a novel Collaborative Knowledge Amalgamation (Co-KA) framework, which aims to preserve both the discriminability and transferability of knowledge in the UKA task. Specifically, within the latent feature space, we introduce the concept of "inherent semantic relationships" between mini-batch samples to maintain knowledge transferability through a distance-based contrastive objective. Additionally, we employ alignment loss to ensure knowledge discrimination between the target student and teachers. In the logit space, we design a decoupled soft-target distillation to enable the student model to acquire both discriminative and transferable knowledge from the aggregate teachers. Extensive experiments on various benchmarks and challenging cross-dataset tasks empirically demonstrate the efficacy and adaptability of our approach for unsupervised knowledge amalgamation. [ABSTRACT FROM AUTHOR]

Feature selection has become a mandatory task in data mining, due to the overwhelming amount of features in Big Data problems. To handle this high-dimensional data and avoid the well-known curse of dimensionality, we need to pre-select an optimal subset of features to reduce redundant computations. Federated learning is a machine learning technique based on training an algorithm over many decentralized edge devices holding local rather than global data on a centralized server. Application of this technique is extending to fields such as self-driving cars, medicine and health, and Industry 4.0, where data privacy is compulsory. Feature selection through federated learning is a complicated task since suboptimal features calculated by feature selection methods may be different in heterogeneous datasets from different nodes. In this paper, we propose a lossless federated version of the classic minimum redundancy maximum relevance (mRMR) feature selection algorithm, called federated mRMR (fed-mRMR), which, without losing any effectiveness of the original mRMR method, is applicable to federated learning approaches and capable of dealing with data that are not independent and identically distributed (non-IID data). Implementation can be found at: https://github.com/jorgehermo9/fed-mrmr • Feature Selection (FS) is a key step in machine learning. • Classical FS does not consider privacy preservation. • Federated FS allows to perform feature selection in federated environments. • fed-mRMR is a federated feature selection tool publicly available. • fed-mRMR performs better than original's mRMR proposal in terms of computation. [ABSTRACT FROM AUTHOR]

This paper introduces an observer-based neural quadratic sliding mode control strategy for interconnected Markov jump systems faced with unknown interconnections, regardless of the high dimensionality of the systems. Firstly, a dynamic event-triggered scheme is constructed in the communication channel to the Lebesgue state observer, with which an integral quadratic sliding mode hyperplane is put forward; Secondly, a neural-based control method is put forward to make sure that predefined sliding hyperplane is attractive; In addition, the occurrence of Zeno phenomenon is also verified to be avoided with the implementation of the controller; Thirdly, linear matrix inequality technique and Lyapunov stochastic stability theory are proposed to check the stochastic stability of closed-loop systems, including sliding mode dynamics and error dynamics; Finally, simulation results on single-link robot arms are given to reveal the validity of the obtained results. [ABSTRACT FROM AUTHOR]

The k Nearest Neighbors (k NN) search in high-dimensional space is a fundamental problem with various applications. In this paper, we try to solve this problem by using deep neural networks (DNNs). We apply DNNs to represent complex correlations between high-dimensional objects, enabling us to project similar objects into the same class, thus reducing the search cost. Based on DNNs, we propose two novel techniques to improve query efficiency while keeping accuracy. First, traditional DNNs typically demand extensive training data for achieving high accuracy. To decrease the training size, we design a multi-module DNN framework comprising several small modules. Each module learns to capture part of knowledge for the given query. The collective output of these sub-modules is then seamlessly integrated to form the final result. Second, with machine learning models, the size of candidates are unbounded. Thus, we design a linear-time data layout refinement algorithm, aiming to limit the number of candidates to a small constant. Empirically we find that our approach significantly outperforms the state-of-the-art methods in terms of both time efficiency and space efficiency while still attaining comparable or better accuracy. [ABSTRACT FROM AUTHOR]

This paper focuses on the design of an adaptive fuzzy filter for discrete-time multi-area power systems, considering the presence of sojourn probabilities and deception attacks. To capture the dynamic behavior of multi-area power systems more accurately, an innovative switching law is introduced, incorporating nonhomogeneous sojourn probability information. The feedback sojourn probability information is determined by a deterministic switched signal, enhancing the system's stability and performance. With the aim of addressing the potential impact of measurement outlier, a robust filter is developed, incorporating a dynamic saturation constraint and fuzzy logic systems. The filter effectively mitigates the influence of outliers by adaptively adjusting the saturation level based on estimation errors. The stability analysis is carried out using Lyapunov theory and average dwell-time information, demonstrating the mean square exponentially ultimately bounded of the augmented system. To validate the proposed algorithm, a simulation example is presented, showcasing its effectiveness and practical applicability in real-world scenarios. [ABSTRACT FROM AUTHOR]

The hardly dominated boundary (HDB) is a common feature of multi-objective optimization problems (MOPs). Previous studies have proposed several multi-objective evolutionary algorithms (MOEAs) to deal with the problem characterized by HDBs. Nevertheless, these methods lack consideration of the evolutionary status, potentially resulting in low efficiency when addressing problems that entail a confluence of HDB and other intricate characteristics. This paper proposes an MOEA with evolutionary-status-driven environmental selection (MOEA-ESD) to address such an issue. Specifically, in each generation, the current and historical information are used to evaluate the evolutionary status. Based on the estimated evolutionary status, the proposed MOEA-ESD algorithm adaptively uses three types of environmental selection: the traditional environmental selection (TES) based on Pareto-dominance and crowding distance, a convergence-first environmental selection (CFES) based on Gaussian mixture model clustering, and a diversity-first environmental selection (DFES) based on outlier detection and extreme solutions. In this way, inferior solutions situated on the HDB can be effectively eliminated, thereby facilitating the convergence of the population while concurrently preserving a commendable degree of diversity. Moreover, a set of new benchmark problems with different objective magnitudes and complicated Pareto sets is developed to enrich the features of HDB-MOPs and to verify the algorithm's performance. Our experimental results on 22 HDB-MOPs show the promising performance of the proposed algorithm. The source code of MOEA-ESD is available at https://github.com/CIAM-Group/EvolutionaryAlgorithm_Codes/tree/main/MOEA-ESD. [ABSTRACT FROM AUTHOR]

Palmprint recognition techniques have been widely applied in security authentication. As a typical image processing method, convolutional neural network (CNN) has been applied to extract features from palmprint images. However, most existing CNN methods cannot meet the demand in recognition speed due to their high computational complexity. This paper proposes a simple yet effective network (SYEnet) for palmprint recognition, which is a lightweight neural network. SYEnet is composed of four learnable feature extraction networks, which can extract discriminative information from 9 patches. In addition, a dedicated network for palmprint ROI extraction (PREnet) is designed. PREnet extracts ROI from the original image, generating ROI images of consistent dimensions, which can be adapted to complex environments without limitations of lighting, gesture, image quality, and resolution. Moreover, a new contactless palmprint database is collected to validate our method. The database contains 12,800 images from 400 palms and provides data supplementation for current research on palmprint recognition. The proposed SYEnet is evaluated on eight real-world palmprint databases, in comparison to other related methods. Experimental results prove that our method is more efficient and has higher recognition accuracy. [ABSTRACT FROM AUTHOR]

Three-way decision, an outstanding method to handle decision-making uncertainties, relies on essentially the loss functions derived from the Bayesian risk decision process. Actually, there are plentiful loss functions that depend on the subjective judgment of decision-makers under different decision scenarios, lacking uniform and objective measurement frameworks. This study pays attention to the real clinical diagnosis, and constructs a weighted probability kernel multi-granularity three-way decision method (WKMG-TWD) integrating gray relation analysis (GRA) over a multi-source heterogeneous decision information system (MHDIS). The method establishes a standardized data-driven calculation framework of loss functions. Foremost, the multi-kernel probabilistic similarity is defined and granularity's weights with knowledge consistency are explored. Subsequently, a weighted probability kernel multi-granularity rough set (WKMGRS) is constructed in this paper. Secondly, to introduce the three-way decision, this study proposes the cost-sensitive individual loss functions considering the correlation determined by GRA between decision objects and different decision classes. Ultimately, this study establishes and applies a three-way iterative classification model to hypertension diagnosis. The experimental results confirm the effectiveness and superiority of the model. The main contribution of this paper is twofold. One is to offer a uniform calculation framework for loss functions and granularity's weights. The other is to furnish invaluable guidance for solving complex medical decision-making problems. [ABSTRACT FROM AUTHOR]

How to control information exchange among different users is an important problem in networked systems with many users/agents. Generally speaking, there are several considerations in control of information exchange in a networked system, including (1) to ensure a friend user has sufficient information to perform its tasks, (2) to deprive an adversary user its information to perform its tasks, (3) to minimize information exchange among friend users so that the risk of information leaking is minimized, and (4) to maximize information broadcasted to all users to achieve maximum transparency. In this paper, we investigate the information control problems in the framework of discrete event systems. Based on the problem at hand, we divide users in a networked system into two or more groups. Users in the same group are consider as friends and users in a different group are consider as adversaries. Several information control problems are investigated and solved using a systematic and rigorous approach. Methods are developed to design controllers that send minimum information to its friends to help them to perform their tasks and broadcast maximum information without helping its adversaries. [ABSTRACT FROM AUTHOR]

Large-scale Multi-Agent Reinforcement Learning (MARL) is fundamentally a challenge due to the curse of dimensionality. In a homogeneous multi-agent setting, mean field theory gives an effective way of scalable MARL by abstracting other agents to a virtual mean agent, assuming that the influence between agents is equal and infinitesimal. However, in some real scenarios, only several neighboring agents, rather than all agents, affect the decision-making of an agent, and different neighboring agents may have varying degrees of influence on the agent's decision-making. In this paper, not restricted to a homogeneous setting, we propose adaptive mean field MARL, which is based on the attention mechanism and can be used to deal with many-agent scenarios where there may be different influence relationships among agents. Specifically, we first derive the mean field approximation with adaptive weight and give the error bound of the approximation. Then, we propose adaptive mean field Q-Learning and describe how to obtain the adaptive weight. In addition, we discuss the differences between the proposed approach and existing mean-field MARL methods. Finally, we conduct experiments on simulation platforms, and the results show that the performance of the proposed approach outperforms that of the state-of-the-art method. [ABSTRACT FROM AUTHOR]

Cloud-assisted biometric identification can benefit data owners with limited resources by offloading their storage and identification tasks to cloud servers with abundant resources. However, the adoption of this computing paradigm is hindered by security and privacy concerns. To solve the dilemma, several privacy-preserving cloud-assisted biometric identification protocols have been proposed. Nevertheless, the existing designs suffer from one or more of the following restrictions: (1) support only for data owner-online query, (2) low security level, and (3) inefficiency. Motivated by these challenges, this paper aims to explore a more secure and efficient construction. Concretely, we initialize the error weighted hashing (EWH)-based data packing technique to replace the time-consuming linear scan search during the identification process. Additionally, to ensure compatibility with the EWH-based search algorithm, we formulate an identity (id)-based scoring strategy and introduce a random split-based data encryption algorithm, which circumvents the costly homomorphic encryption scheme. Within the framework of the two non-colluding servers model, the random split-based one-time pad encryption guarantees provably indistinguishable security against chosen-plaintext attacks. Finally, since only simple hash and xor operations are involved, our novel construction exhibits significantly improved performance compared to prior designs, both in theory and in practice. This superiority is validated through rigorous theoretical analysis and extensive experimental evaluation. [ABSTRACT FROM AUTHOR]

Sequential recommendation captures users' dynamic preferences by modeling the sequential information of their behaviors. However, most existing works only focus on users' behavior sequences in a single domain, and when there is insufficient data in the target domain, the recommendation performance may not be satisfactory. We notice that a user's interests are usually diverse, for which the items he/her interacts with in a period of time may be from multiple domains. Moreover, there are also item transition patterns across sequences from different domains, which means that a user's interaction in one domain may affect his/her interaction in the other domains next time. In this paper, we aim to improve the performance of sequential recommendation in the target domain by introducing users' behavior sequences from multiple source domains, and propose a novel solution named transfer via joint attentive preference learning (TJAPL). Specifically, we tackle the studied problem from the perspective of transfer learning and attentive preference learning (APL). For target-domain APL, we adopt the self-attention mechanism to capture the users' dynamic preferences in the target domain. Furthermore, to address the scarcity challenge posed by limited target-domain data, we introduce users' behavioral sequences in the source domain, and devise cross-domain user APL to transfer and share the users' overall preferences from multiple source domains to the target domain. We also design cross-domain local APL that specializes in capturing the item transition patterns across different domains for knowledge transfer. These modules are all based on the attention mechanism and thus can accelerate the training by parallel computation. Notice that our TJAPL can be applied to scenarios with multiple source domains, while transferring knowledge from multiple domains is potentially helpful in practical applications. Extensive empirical studies indicate that our TJAPL significantly outperforms thirteen recent and competitive baselines. [ABSTRACT FROM AUTHOR]

This paper proposes a variable-type hierarchical fuzzy system (VTHFS) with high accuracy and strong interpretability. An improved rear-part direct-connected structure is used in the VTHFS, which eliminates the intermediate variables of the hierarchical fuzzy system (HFS). A new feature sorting method is proposed to find suitable input variables for each layer of subsystems of VTHFS. Under the proposed topology structure, layer by layer training of VTHFS can be achieved by changing the expected output of each layer, thereby reducing residual errors layer by layer and improving the prediction accuracy of VTHFS. Three fuzzy set distribution constraints are used in parameter optimization of VTHFS to improve the interpretability, including integrity, distinguishability, and footprint of uncertainty. The type of fuzzy set is automatically adjusted according to the uncertainty characteristics of input data measured by fuzziness, which further enhances the semantic interpretability of the VTHFS and solves the problem of requiring prior knowledge to define the fuzzy set type in existing research. Finally, the proposed VTHFS is tested on the regression data sets in the real world. Experimental results demonstrate the effectiveness of the feature sorting method and the fuzzy set type adjustment method. Through a large number of quantitative indicators, it can be confirmed that VTHFS has stronger interpretability and higher prediction accuracy compared to other fuzzy systems. [ABSTRACT FROM AUTHOR]

• Analyze DE algorithm two key issues: parameter sensitivity and search imbalance. • A complementary mutation strategy based on complementary symmetric selection and a Gaussian distribution mutation is proposed. • Data fusion-based parameter adaptation method proposed for parameter rationalization. • A novel variant of DE called DACDE is designed and results shown competitive edge. As an excellent optimization algorithm widely used to solve various practical problems, differential evolution (DE) algorithm has few parameters, yet its performance is significantly affected by these parameters. To address this issue, this paper presents a novel variant of DE called DACDE, which utilizes data fusion-based parameter adaptation and a complementary mutation strategy. Most parameter adaptation methods based on successful history only analyze the mean of parameters and ignore their degree of dispersion. In DACDE, the successful parameter distribution is recorded and described by both the mean and variance. Data fusion is then used to combine records and generate an estimated distribution, which is applied to a Gaussian distribution to generate new parameters. Inspired by opposition-based learning, we introduce a complementary mutation strategy. This strategy employs a symmetric selection mechanism to adapt to the varying search abilities required by the algorithm at different stages. The new variant is verified on 32 single-objective functions from CEC 2011 and 2014 benchmark suites, and the results show that DACDE is competitive compared to other 29 evolutionary algorithms. [ABSTRACT FROM AUTHOR]

Aspect-based sentiment classification has become a popular topic in natural language processing. Exploiting dependency syntactic information with graph neural networks has recently become a popular trend. Despite their success, methods that rely heavily on a dependency tree face major challenges. This concerns the alignment of aspects and their word sentiments due to the richness of the language and the fact that a dependency tree might produce noisy signals from unrelated associations. This paper introduces a Dual-Relational Graph Attention Network (DRGAT) that fully exploits syntactic structural information and then the sentiment-aware context (e.g., phrase segmentation and hierarchical structure) of the constituent tree of a sentence. Additional constituency and dependency attention mechanisms provide comprehensive syntactic information across words, thereby enabling an accurate connection between aspect words and corresponding sentiment words. Considering that the original parsed constituency tree may have a large depth, this could lead to words being far apart increasing the computational overhead. The constituency tree of each sentence is dynamically reconstructed by determining the importance of each relational node. Extensive experimental results on six English datasets demonstrated that fully exploiting syntactic information can achieve excellent sentiment classification results. [ABSTRACT FROM AUTHOR]

Histogram equalization (HE) is a well-established method for image contrast enhancement due to its simplicity and effectiveness. However, it suffers from three main shortcomings, i.e., over-enhancement, under-enhancement and mean shift. To address these issues, this paper proposes a systematic scheme, that is, adaptive histogram equalization with visual perception consistency (AHEVPC). Firstly, a novel histogram correction model is designed to get the optimal controlling parameters, which specifically address the aforementioned issues of HE. Besides, considering the subjective perception of the initial output of the model, two strategies are proposed to make the enhanced image more natural and comfortable. Finally, histogram equalization is applied to the modified histogram. Extensive experimental results demonstrate that the proposed scheme is reasonable and effective, and outperforms several state-of-the-art methods in terms of subjective and objective metrics. [ABSTRACT FROM AUTHOR]

In a modern distributed learning framework, the speeds of intra-worker calculation and inter-worker communication may differ by 1,000 times. It is advisable to perform expensive calculations on worker machines and communicate as few rounds as possible. In this paper, we propose three novel distributed sparse learning algorithms (EDSL-ET, EDSL-IBCD, and EDSL-IBCD-ET) in high dimensions, which are efficient in both computation and communication. Algorithm EDSL-ET utilizes the Top k sparse technique to reduce communication costs and adopts error feedback to guarantee convergence. Algorithm EDSL-IBCD greatly reduces communication costs by introducing an independent block coordinate descent method. Algorithm EDSL-IBCD-ET takes advantage of EDSL-ET and EDSL-IBCD algorithms for extra-high dimensional feature learning, which is the best one whether in communication, computation, and stability. We give theoretical guarantees of the three algorithms under mild conditions, which match the performance of the centralized algorithm. Extensive experiments on simulated and real data validate our theoretical analysis and demonstrate that the proposed algorithms perform well with just a few rounds of communications. [ABSTRACT FROM AUTHOR]

Robust deadlock supervisory control techniques for automated manufacturing systems under resource failures that need additional central buffers may lead to supererogatory cost in control implementation. To mitigate this issue, this paper reports a Petri net-based low-cost adaptive supervisory control policy that does not require extra buffers. If an unreliable resource fails, three classes of buffer spaces can be temporarily used to store parts that require the failed resource in their impending processing routes. The proposed adaptive supervisory control comprises of control places and switch controllers. If an unreliable resource fails, the switch controllers are activated to move the part types that require the failed resource in their subsequent processing stages into borrowed buffer spaces. After the failed resource is recovered, the system returns to its normal operating mode. The part types in the borrowed buffer spaces will be returned to their last processing stages before the occurrence of the failure and continued along their processing route. We propose criteria for selecting buffer space borrowers and buffer space lenders. Furthermore, redundant control places are removed by a new technique proposed in the paper in order to reduce the complexity of the supervisory control structure. We demonstrate the proposed method using examples. [ABSTRACT FROM AUTHOR]

This paper presents a comprehensive empirical investigation into the interactions between various randomization techniques in Deep Neural Networks (DNNs) and their impact on learning performance. It is well-established that injecting randomness into the training process of DNNs, through various approaches, at different stages, is often beneficial for reducing overfitting and improving generalization. Nonetheless, the interactions between randomness techniques such as weight noise, dropout, and many others remain poorly understood. Consequently, it is challenging to determine which methods can be effectively combined to optimize DNN performance. To address this issue, we categorize the existing randomness techniques into four key types: injection of noise/randomness at the data, model structure, optimization or learning stage. We use this classification to identify gaps in the current coverage of potential mechanisms for the introduction of randomness, leading to proposing two new techniques: adding noise to the loss function and random masking of the gradient updates. In our empirical study, we employ a Particle Swarm Optimizer (PSO) for hyperparameter optimization (HPO) to explore the space of possible configurations to determine where and how much randomness should be injected to maximize DNN performance. We assess the impact of various types and levels of randomness for DNN architectures across standard computer vision benchmarks: MNIST, FASHION-MNIST, CIFAR10, and CIFAR100. Across more than 30 000 evaluated configurations, we perform a detailed examination of the interactions between randomness techniques and their combined impact on DNN performance. Our findings reveal that randomness through data augmentation and in weight initialization are the main contributors to performance improvement. Additionally, correlation analysis demonstrates that different optimizers, such as Adam and Gradient Descent with Momentum, prefer distinct types of randomization during the training process. A GitHub repository with the complete implementation and generated dataset is available.2 [ABSTRACT FROM AUTHOR]

The paper presents a new balanced selection operator applied to the proposed Balanced Non-dominated Tournament Genetic Algorithm (B-NTGA) that actively uses archive to solve multi- and many-objective NP-hard combinatorial optimization problems with constraints. The primary motivation is to make B-NTGA more efficient in exploring Pareto Front Approximation (PFa), focusing on "gaps" and reducing some PFa regions' sampling too frequently. Such a balancing mechanism allows B-NTGA to be more adaptive and focus on less explored PFa regions. The proposed B-NTGA is investigated on two benchmark multi- and many-objective optimization real-world problems, like Thief Traveling Problem and Multi-Skill Resource-Constrained Project Scheduling Problem. The results of experiments show that B-NTGA has a higher efficiency and better performance than state-of-the-art methods. • New balanced GAP selection actively uses archive, focuses on Pareto Front 'gaps' to reduce too frequent sampling of regions. • B-NTGA bases and modifies NTGA2 (i.e., clone elimination, tournament as second selection) and adds balanced GAP selection. • B-NTGA is less time-consuming than NTGA2, uses fewer parameters, and is more efficient than NTGA2. • Experiments for NP-hard multi-objective optimization problems (TTP and MS-RCPSP) comparing B-NTGA to state-of-the-art methods. • Results of experiments showed that B-NTGA is highly effective for multi- and many-objective problems (like TTP and MS-RCPSP). [ABSTRACT FROM AUTHOR]

• A hybrid solution that uses a genetic algorithm and models property features as fuzzy partitions is proposed for selecting the residential properties whose economic value is most consistent with the purchase offer. • Our framework does not need a learning phase such as in Machine Learning real estate price assessment models. • Expert knowledge is modeled using fuzzy partitioning of the feature domains and evaluating positive or negative deviation value from the average unit price. • The reliability of the solution is evaluated by measuring its fuzziness. Evaluating the actual price of a residential property is a critical issue in the real estate market. Real estate market practitioners gauge a property's price by considering features such as property type and residential area. Subsequently, they evaluate the property's intrinsic features, such as condition, sun exposure, scenic views, and ancillary amenities. Finally, extrinsic features such as the proximity of services and infrastructure are assessed. This paper proposes a new genetic approach for selecting residential properties that meet the purchase offer and the intrinsic and extrinsic characteristics desired by the client. Since the real estate market's changes can influence extrinsic features, the method introduces price fluctuations of properties. Extrinsic features are modelled as fuzzy partitions: each fuzzy set describes a qualitative aspect of the corresponding feature that, expressed in a linguistic term, has a human-like interpretation. Then, a deviation value (fluctuation) from the average price of the property is considered for each fuzzy set in the partition. All the property features, extrinsic and intrinsic, are encoded in the chromosome genes of the genetic algorithm. The fitness function calculates the distance between the unit price of the property and the purchase offer. Some case studies were conducted in various Italian municipalities, using the average price per square meter of residential properties the Osservatorio del Mercato Immobiliare (OMI) assigned. Depending on customer requirements and preferences, different OMI zones were selected using additional characteristics such as type, location, conservation, and proximity to various urban services. The results demonstrated the effectiveness of the proposed approach for all the case studies, showing how the optimal solution represents a good compromise between customer preferences and market offerings. [ABSTRACT FROM AUTHOR]

Federated learning (FL) enables the full utilization of decentralized training without raw data. However, various attacks still threaten the training process of FL. To address these concerns, differential privacy (DP) and secure multi-party computation (SMC) are applied, but these methods may result in low accuracy and heavy training load. Moreover, the high communication consumption of FL in resource-constrained devices is also a challenging problem. In this paper, we propose a novel SMC algorithm for the FL (FL-IPFE) to protect the local gradients. It does not require a trusted third party (TTP) and is more suitable for FL. Furthermore, we propose a secure and efficient FL algorithm (SEFL), which applies compressed sensing (CS) and all-or-nothing transform (AONT) to minimize the number of transmitted and encrypted model updates. Additionally, our FL-IPFE is used to encrypt the last element of the preprocessed parameters for guaranteeing the security of the entire local model updates. Meanwhile, the issue of participant dropouts is also taken into account. Theoretical analyses demonstrate that our proposed algorithms can aggregate model updates with high security. Finally, experimental evaluation reveals that our SEFL possesses higher efficiency compared to other state-of-the-art works, while providing comparable model accuracy and strong privacy guarantees. [ABSTRACT FROM AUTHOR]

Information fusion technology plays a crucial role in integrating data from multiple sources or sensors to generate comprehensive representation, which can eliminate uncertainty in multi-source information systems (Ms-IS). Incomplete interval-valued data, a generalized form of single-valued data, is commonly encountered in real-world scenarios and effectively represents uncertain information. This paper introduces a novel information entropy specifically designed to quantify the uncertainty in incomplete interval-valued data. Based on the proposed entropy, a new unsupervised fusion approach is developed. Additionally, two dynamic update mechanisms are established to obtain fusion results efficiently when collecting new objects and removing obsolete ones. The relevant static and dynamic fusion algorithms are provided, and a detailed analysis and comparison of their time complexities are conducted. Finally, the effectiveness analysis reveals that the proposed method achieves higher average classification accuracy (5% to 8.7% improvement) compared to three common fusion methods (MAXF, MEANF, and MINF), as well as the state-of-the-art entropy-based supervised fusion method (ESF). The efficiency analysis demonstrates that the average running time of dynamic fusion algorithms is significantly lower (66.9% to 85.6% reduction) compared to the static fusion algorithm, and this difference is statistically significant. [ABSTRACT FROM AUTHOR]

Recently, graph data become increasingly important and larger in many fields, and many distributed graph computing systems have been proposed to deal with large-scale graphs. The graph partition algorithm is the basis of these systems. In many graph applications, a vertex updates its feature by aggregating information of its multi-order neighboring vertices, and the existing graph partition algorithms often lead to heavy communication cost. This paper proposes a graph partition algorithm to store the multi-order neighbor vertices redundantly to avoid the communication requirement. It first formulates the problem of graph partitioning with redundant multi-order neighbor vertices storage as an optimization problem, and then proposes PARN (partition algorithm with redundant multi-order neighbor) algorithm to tackle it. The PARN algorithm consists of initial partition and vertex migration based on genetic algorithm, and it classifies the vertices of a partition into primary and auxiliary vertices. The auxiliary vertices required for the primary vertices in a partition are stored in the same partition, so the primary vertices don't need to fetch the information from the other partitions. The experiment results show that the proposed algorithm is comparable to Hash, GAP, GNP, and QCLP algorithms in terms of load balance and produces fewer auxiliary vertices than these algorithms. • Formulating the graph partitioning problem with redundant storage of multi-order neighbor vertices. • Presenting a graph partitioning algorithm PARN to solve this problem. • Performing extensive experiments to show the out-performance of the proposed algorithm. [ABSTRACT FROM AUTHOR]

With the development of the Internet and social media, rumors can spread not only through word-of-mouth but also rapidly through the network. In this paper, a dynamic model of rumor propagation with time delay is proposed separately for non-network and network scenarios. Additionally, we analyze the equilibrium points and their existence conditions for rumor propagation. After the linear approximation of the model, the necessary conditions for Turing instability are derived. Furthermore, the amplitude equation corresponding to the model in this paper is derived. Finally, through numerical simulations in the non-network model, we validate the aforementioned theories and find that changing the removed coefficient and cross-diffusion coefficient have a significant impact on Turing patterns, while time delay and periodic diffusion have a smaller impact. In the network model, we compare the effects of two network structures, the WS network and the BA network, through numerical simulations, verifying the feasibility of the Monte Carlo simulation method. [ABSTRACT FROM AUTHOR]

Hybrid learning is an excellent method that combines the global information of data with the local information of data. Different from the known hybrid learning algorithms, in this paper we propose a new hybrid learning strategy for Fisher Linear Discriminant (FLD) and introduce novel hybrid learning based on FLD (HL-FLD). The main idea of HL-FLD is to obtain the global structural information by FLD firstly, and then use the obtained global information to divide the given data locally in a more detailed way. To study systematically the proposed HL-FLD, we not only establish the generalization bound of HL-FLD and prove that the proposed HL-FLD algorithm is consistent, but also present some discussions on HL-FLD. Since splitting the blocks of a given data is a hyperparameter, we also improve HL-FLD and introduce another new self-adaptive hybrid learning based on FLD (SHL-FLD). The experimental researches for benchmark repository confirm that the proposed two algorithms have better performance in terms of misclassification rates and total time. [ABSTRACT FROM AUTHOR]

With the rapid development of e-commerce, more and more people are willing to post their reviews and opinions about the products they buy online. Therefore, the use of online review data for decision-making appears to be more practical and universal, and how to effectively use this kind of data to support large-scale group decision-making (LSGDM) is a worthy research direction. In this paper, we firstly use sentiment analysis to analyze online review data to derive the decision maker's (DM's) sentiment value, and apply the sentiment value to construct a social network based on a dual trust relationship, which considers both familiarity-based trust and similarity-based trust. Secondly, a directed Louvain clustering algorithm in light of dual trust relationships and a method for solving the DM's intra-group weights and the group's weights are proposed based on this network. A two-stage clustering based consensus model in light of dual trust relationships is then proposed, in which the DMs in the agreement cluster can communicate with other DMs outside of such a cluster and dynamically update the grouping using the clustering algorithm. Finally, the practicality and effectiveness of the LSGDM method proposed in this paper are verified through a real case. [ABSTRACT FROM AUTHOR]

Heterogeneous graphs (HGs) contain many nodes and their interaction relationships, which can model complex systems and provide rich semantic and structural information for task execution. Among these, HG representation stands as the fundamental and pivotal component. Existing HG representation methods primarily employ graph neural networks to acquire the semantics of nodes along various meta-paths and fuse them to represent the nodes. The most prevalent HG representation methods encompass two steps: semantic information extraction within meta-paths and semantic fusion between meta-paths. However, these methods overlooked the consideration of node heterogeneity within meta-paths and the simultaneous semantic correlation between meta-paths. Specifically, node heterogeneity within meta-paths signifies that the meta-path-based neighbors do not consistently contain information that positively influences the target node, and the semantic correlation between meta-paths indicates that different meta-path spaces are not entirely independent. Disregarding either of these issues leads to the propagation of irrelevant or redundant information and potential disruption of HG embedding. Consequently, in this study, we propose the HBHG, which is a hierarchical bottleneck for heterogeneous graph representation. HBHG primarily employs the information bottleneck (IB) as a guiding principle, constraining the propagation of irrelevant information within and between meta-paths while preserving relevant information. The central concept of the IB revolves around viewing model learning as the preservation of relevant information and compression of irrelevant information, accomplished by minimizing the dependency between input and hidden features through mutual information (MI) and maximizing the dependency between hidden features and ground-truth. Considering the complexity associated with MI estimation, this paper introduces a novel dependency index, namely the Hilbert-Schmidt independence criterion (HSIC), which offers ease of calculation. Specifically, HBHG comprises two primary components: a semantic bottleneck within meta-paths and a semantic bottleneck between meta-paths. The semantic bottleneck within meta-paths relies primarily on the HSIC-based limitations of dependencies at different layers of the graph neural network on various meta-paths, thereby maximizing the extraction of information relevant to the target node from neighboring nodes. The semantic bottleneck between meta-paths enables flexible extraction and fusion of semantic information based on downstream tasks, achieved by managing the trade-off of dependencies with HSIC between different meta-path semantic spaces. In summary, the proposed HBHG integrates hierarchical bottleneck constraints within and between meta-paths. This integration serves to maximize the aggregation of relevant information while effectively compressing irrelevant information, thereby enhancing the quality of heterogeneous graph embedding. The effectiveness of HBHG was validated through performance and ablation experiments conducted on multiple datasets. [ABSTRACT FROM AUTHOR]

Pseudo-labeling-based approaches are gaining prominence in Semi-Supervised Learning (SSL). Recent studies have identified that the key bottleneck in this methodology is addressing insufficient, incorrect, and imbalanced pseudo-labels. In this paper, we argue that the intrinsic problem behind this bottleneck is classifier bias, i.e., the classifier's prototypes suffer from poor uniformity. Further, inspired by neural collapse that reveals an optimal structure under supervised training scenarios, we address this classifier bias by utilizing an offline simplex Equiangular Tight Frame (ETF) classifier with maximally and equally separated prototypes. During the training phase, we maintain the prototypes of the classifier as fixed and concentrate on refining the feature encoder. Specifically, we integrate a straightforward clustering-based pseudo-labeling strategy with information maximization for feature learning. In practice, the fixed ETF classifier prevents the model from falling into a detrimental cycle, where a biased classifier exacerbates misaligned features, further perpetuating this bias. Furthermore, the clustering-based pseudo-labeling strategy reduces the dependency on complex threshold-adjusting mechanisms and effectively navigates the quantity-quality trade-off that plagues existing SSL methods. Leveraging these methodologies, we develop a simple yet powerful approach, termed ETF-SSL. Extensive experiments across Image, Text, and Audio datasets demonstrate that ETF-SSL can achieve competitive or superior performance compared to existing approaches. This success highlights the benefits of using a fixed ETF classifier in SSL and points to promising directions for future research in this area. The code is available at: https://github.com/yichenwang231/ETFSSL. [ABSTRACT FROM AUTHOR]

Stochastic configuration networks (SCNs) are a class of randomized learner models that have garnered increasing attention in data analytics. In the original implementation of SCN, the pseudo-inverse of the hidden layer output matrix is used to evaluate output weights. However, it is quite challenging to compute the pseudo-inverse of the hidden layer output matrices for large-scale datasets with high complexity, which occurs in many real-world applications. This paper aims to accelerate the learning process of SCNs by employing the well-known alternating direction method of multipliers (ADMM) solver and CPU-GPU heterogeneous computing technique. Empirical results on two benchmark datasets demonstrate the efficiency and effectiveness of the proposed method for large-scale data analytics. [ABSTRACT FROM AUTHOR]

Reverse Approximate Nearest Neighbour (RANN) query relaxes the R k NN definition of influence, where a user u can be influenced by not only its closest facility but also by every other facility that is almost as close to u as its closest facility is. In this paper, we study the continuous monitoring of RANN queries on road network. Existing continuous RANN algorithms on Euclidean space cannot be extended to continuously monitor RANN queries on road network. We propose two different methods to efficiently monitor RANN queries. We conduct an extensive experiment on different real data sets and demonstrate that our both proposed algorithms are significantly better than the competitor [ABSTRACT FROM AUTHOR]

Temporal Knowledge Graphs (TKGs) model time-dependent facts as relations between entities at specific timestamps, making them well-suited for real-world scenarios. However, TKGs are susceptible to incompleteness, necessitating Temporal Knowledge Graph Completion (TKGC) to predict missing facts. Prior methods often struggle to effectively handle two critical properties of TKGs, time-variability and time-stability, simultaneously, which hinders their performance. In this paper, we propose Space Adaptation Network (SANe), a novel approach for TKGC. SANe adapts facts at different timestamps to distinct latent spaces, effectively addressing time-variability. Our model introduces Parameter Generation Network to produce separate neural networks for each snapshot, which are then encoded into different latent spaces. A dynamic convolutional neural network processes entities and relations, utilizing different learned parameters generated by parameter generation network with respect to timestamps. By handling different temporal snapshots separately, TKGC is transformed into static KGC, enabling the modeling of time-variability. Dynamic convolutional neural network efficiently learns collective knowledge over large periods and supplements more specific knowledge gradually in smaller periods, facilitating time-stability. To strike a balance between learning time-variability and time-stability, we introduce a time-aware parameter generator to produce parameters hierarchically based on year, month, and day timestamps. Long-term knowledge is effectively shared across adjacent snapshots within the same year or month, while short-term knowledge within a day is preserved in specific parameters. However, in unbalanced TKGs, where many facts occur in small intervals, the large number of parameters generated by time-aware parameter generator may remain underutilized. To address this, we propose Adaptive Parameter Generation with a partition tree, ensuring parameter load balancing while maintaining time-stability. We conduct extensive experiments on five benchmark datasets, demonstrating the superiority of SANe over existing methods for TKGC, achieving state-of-the-art performance. Our contributions include pioneering TKGC from the perspective of space adaptation, achieving a balance between time-variability and time-stability through latent space overlap constraints, and substantiating the effectiveness of our model through comprehensive experiments on rich temporal datasets. [ABSTRACT FROM AUTHOR]

A rule-based system is a widely used artificial intelligence system that employs a set of rules to make decisions. The belief rule-based (BRB) classification system is an extension of fuzzy rule-based (FRB) system that handles uncertainty and imprecision in classification tasks by incorporating Dempster-Shafer evidence theory and fuzzy set theory. However, the BRB classification system suffers from the combinatorial explosion and high time complexity problems. To solve these issues, the fuzzy unordered rule induction algorithm (FURIA) is introduced into the BRB classification systems to design a novel BRB classification system in this paper. FURIA is computationally less complex and has superior classification performance. The proposed system outperforms BRB classification systems in terms of classification performance and significantly reduces the number of rules and conditions. Furthermore, the proposed system offers a more effective solution than FURIA. To evaluate the validity and superiority of the proposed system, we conduct two classification experiments, comparing it with five traditional classifiers and seven rule-based systems, respectively, in terms of classification performance and interpretability. [ABSTRACT FROM AUTHOR]

Regression methods are widely employed in face recognition applications. The vector-based norm regression model only represents the pixel by pixel correlation and exhibits poor performance with contiguous occlusion. In contrast, methods based on the nuclear norm describe the low-rank structural information, but it is difficult to achieve satisfactory performance when dealing with complex noise. To address this, a log-based non-convex relaxation regularized regression (log-NCRR) model for robust face recognition is proposed in this paper. We adopt the log-based matrix loss without additional parameters to characterize the low-rank part of error image. Considering sparsity, we design a log-based vector loss to describe the sparse part of the error matrix, which achieves a balance between the l 0 -norm and l 1 -norm. Additionally, a weighted non-convex relaxation of l 2 , 1 -norm is proposed to ensure the group sparsity of the regression coefficient. The optimization problem is optimized by ADMM, with the corresponding sub-problems easily solved by the generalized singular value shrinkage operator. Finally, experimental results on four public databases for various types of noise validate the effectiveness and robustness of log-NCRR in face recognition. • The log-based matrix loss is applied to face recognition. It serves as an appropriate approximation of the rank without additional parameters. • We propose the log-based vector loss to approximate l 1 -norm. The non-convex relaxation of the l 2 , 1 -norm ensures group sparsity. • In the process of solving sub-problem of the original problem, the log-based shrinkage operators are introduced in this research. [ABSTRACT FROM AUTHOR]

Space Surveillance Network (SSN) task scheduling plays a crucial role in maintaining the catalog of Resident Space Objects (RSO). However, various emergencies, such as RSO maneuvering, collisions, or rocket launch, may disrupt the original scheduled scheme. Therefore, it is essential to rapidly regenerating emergency schemes while minimizing disturbance to the initial scheduling scheme. This paper introduces an Emergency Task Scheduling model, referred to as MM-ETS, which aims to Maximize observation profits and Minimize disturbance. This model incorporates constraints related to observability, tasks, and resources, which are derived from practical applications. Additionally, a Hierarchical Distributed Dynamic Emergency Scheduling algorithm, encompassing Task assignment, Conflict resolution, Resource negotiation, and Center collaboration (HD-TCRC-DES), is proposed. The presented algorithm is activated by emergencies and a Rolling Horizon Strategy (RHS) is employed to break down long-term, large-scale problems into short-term, small-scale problems, thus improving feasibility and emergency response capabilities. At each layer of the proposed algorithm, heuristic rules are utilized to solve the new scheme, which helps allocate the computational load to resource nodes, and quickly adjust the initial scheduling scheme. Experimental scenarios involving 15 ground observation resources and 1000 emergency tasks are constructed, and the simulation results demonstrate that the proposed method can enhance Comprehensive Benefits Indicators (CBI) by approximately 24.65%, 24.7%, 24.91%, and 30.5% compared to the baselines. Consequently, it is suitable for SSN emergency task scheduling. [ABSTRACT FROM AUTHOR]

Simplicial map neural networks (SMNNs) are topology-based neural networks with interesting properties such as universal approximation ability and robustness to adversarial examples under appropriate conditions. However, SMNNs present some bottlenecks for their possible application in high-dimensional datasets. First, SMNNs have precomputed fixed weight and no SMNN training process has been defined so far, so they lack generalization ability. Second, SMNNs require the construction of a convex polytope surrounding the input dataset. In this paper, we overcome these issues by proposing an SMNN training procedure based on a support subset of the given dataset and replacing the construction of the convex polytope by a method based on projections to a hypersphere. In addition, the explainability capacity of SMNNs and effective implementation are also newly introduced in this paper. [ABSTRACT FROM AUTHOR]

Three-way decision model is an effective way to deal with complex decision problems. However, since the three-way decision models now proposed are all based on a single decision criterion, the decision results typically reflect only one preference of decision-makers. Thus, these models may also not effectively deal with complex decision-making problems. To solve the above problems, this paper proposes a new three-way decision model based on ensemble learning. Specifically, we first obtain different three-way decision results by employing different decision criteria. Then, we can acquire the core and candidate sets of the positive and negative regions through set operations. Next, we use the K-means algorithm to divide the candidate sets into three disjoint subsets based on similarities. After that, we adopt a hierarchical filtering method to select suitable objects from the candidate sets and add them to the core sets. Finally, we employ four three-way decision models with different decision criteria as examples to conduct experiments on eight datasets. Experimental results show that our proposed model can obtain higher classification accuracy and lower deferment rate than other traditional three-way decision models under most experimental conditions. • A new three-way decision model based on ensemble thinking is put forward. • The core and candidate sets for a particular set are given based on set operations. • The HCS , MCS and LCS are constructed by using K-means algorithm. • A hierarchical filtering method is designed to select appropriate objects. [ABSTRACT FROM AUTHOR]

Non-parallel hyperplane classifiers have attracted many attention due to their wider applicability than parallel hyperplane classifiers, and it accelerates the training speed through separately training of each hyperplane. However, it does not consider the correlation relationship among the non-parallel hyperplanes. In addition, since these non-parallel hyperplanes are built on the same training data, there may exist correlation among them due to redundancy of the data. In this paper, we investigate the decorrelation problem of non-parallel hyperplane classifiers. We integrate non-parallel hyperplanes into a unified model and explore this relationship through a joint classifier learning approach. Taking the twin support vector machine as an example, a novel decorrelation regularization term is added into the joint problem. Additionally, an effective alternating optimization algorithm has been introduced to address this nonconvex problem. To evaluate the proposed method, a series of experiments are conducted on binary and multi-class classification datasets from the benchmark database repository. Experimental results compared to several parallel and non-parallel hyperplane based classifiers demonstrate the effectiveness of the proposed method. • A unified non-parallel SVM with decorrelation TWJCL for multi-class classification is proposed. • Compared to JCL, TWJCL achieves decorrelation in each binary classification. • Experiments show TWJCL has low correlation coefficient and good classification performance. [ABSTRACT FROM AUTHOR]

Multi-view clustering (MVC) can exploit the complementary information among multi-view data to achieve the satisfactory performance, thus having extensive potentials for practical applications. Although Nonnegative Matrix Factorization (NMF) has emerged as an effective technique for MVC, the existing NMF-based methods still have two main limitations: 1) They solely focus on the reconstruction of original data, which can be regarded as the decoder of an autoencoder, while neglecting the low-dimensional representation learning. 2) They lack the ability to effectively capture both linear and nonlinear structures of data. To solve these problems, in this paper, we propose a D ual A uto-weighted multi-view clustering model based on A utoencoder-like NMF (DA2NMF), which enables a comprehensive exploration of both linear and nonlinear structures. Specifically, we establish an autoencoder-like NMF model that learns linear low-dimensional representations by integrating data reconstruction and representation learning within a unified framework. Moreover, the adaptive graph learning is introduced to explore the nonlinear structures in data. We further design a dual auto-weighted strategy to adaptively compute weights for different views and low-dimensional representations, thereby obtaining an enhanced consistent graph. An effective algorithm based on Multiplicative Update Rule (MUR) is developed to solve the DA2NMF with the theoretical convergence guarantee. Experimental results show that the proposed DA2NMF can effectively improve the clustering performance compared with the state-of-the-art MVC algorithms. [ABSTRACT FROM AUTHOR]

Co-clustering aims to cluster the rows and columns of data simultaneously and can be often formulated as a two-objective optimization problem (one objective for rows and the other for columns) and the solution is a Pareto-optimal solution set in principle. Existing methods usually convert the co-clustering problem into a single-objective optimization problem by setting a hyper-parameter λ between the two objectives. However, finding a good value of λ is not easy because of its large parameter space; also, there may exist many equally good λ s. In this paper, a nondominated sorting genetic model (NSGC) is proposed to tackle the co-clustering problem, totally bypassing the trade-off parameter λ and returning to the original two-objective problem. The core of our model is to group a row objective function and a column objective function and integrate them into a genetic algorithm as the fitness functions. After this reformulation, we follow a standard genetic algorithm procedure to iteratively find the Pareto-optimal solutions. Finally, to fish out a single best solution we further design a sorting criterion according to which the Pareto-optimal solution set can be totally ordered. Extensive experiments with 16 public datasets are conducted, and the results demonstrate the superiority of our approach. [ABSTRACT FROM AUTHOR]

As a classical unsupervised learning method, the K-means algorithm selects the cluster centers randomly and calculates the mean values of the cluster's data points to generate clusters. However, its performance is susceptible to the initial cluster centers and the sparsity of the membership matrix. To overcome these limitations, in this paper, we propose a sparse K-means clustering algorithm with anchor graph regularization (SKM-AGR) for optimizing initial cluster center sensitivity and improving membership matrix sparsity. The main idea is to use the anchor graph regularization (AGR) constrained K-means models, which effectively learn the membership matrix of data points and the membership matrix of anchors. In particular, by constructing an anchor graph, the AGR term not only discovers the internal structure information of data, but also covers the data distribution. Furthermore, an alternating optimization algorithm with fast-converging is adopted to solve the optimization problems of SKM-AGR, and the computational complexity is analyzed. Extensive clustering experiments on several synthetic and benchmark datasets show that the proposed SKM-AGR method performs better than several previous methods in most cases. [ABSTRACT FROM AUTHOR]

Synthetic Aperture Radar (SAR) images are widely used in ship detection because of their all-weather and all-day imaging characteristics. However, there are two challenges for SAR ship detection. One is coherent speckle noise, causing ship confusion with similar objects and raising false alarms. The other is multi-scale ship detection, particularly in small ships, which suffers from insufficient accuracy. To address these challenges, this paper proposes a robust one-stage detector, S3MDet, for SAR ship detection with sequential three-way decisions (S3WDs) and multi-granularity. First, to effectively eliminate the interference of coherent speckle noise, a noise classification and denoising module (NCDM) S3WD-based is designed. This module can accurately classify the noise level of the image and only perform denoising on the images identified as noisy, avoiding unnecessary operations on noise-free images. Then, to solve the problem of multi-scale ship detection, a multi-granularity group attention module (MGAM) is designed to obtain a richer representation of multi-granularity features. This module adopts a multi-granularity group convolution structure and channel-wise attention weights to efficiently extract ship features of different scales from SAR images. Extensive experiments on four SAR ship datasets, including SAR-Ship-Dataset, HRSID, SSDD, and LS-SSDD-v1.0, validate the robustness of S3MDet, demonstrating that it achieves state-of-the-art performance. [ABSTRACT FROM AUTHOR]

With Community Question Answering (CQA) sites evolving into quite popular knowledge-sharing platforms on the Internet, they have also become ideal places for various spammers to spread fake or promotional information. Recently, with the rapid development of crowdsourcing systems, numerous malicious users have launched organized spam campaigns, conducting many spam accounts to carry out collusive spamming activities on CQA sites. In these campaigns, the spammers do not act independently but post deceptive questions and answers (Q&As) collaboratively, which makes the Q&As closely related to each other, but the spam clues of them are even less visible. Therefore, most existing spam detection works may fail to detect these carefully organized and posted collusive CQA spam. In this paper, taking Baidu Zhidao, a popular CQA platform in Chinese, as the study object, we propose a C ollective C lassification framework for community Q uestion A nswering spam detection (CCQA), which collectively identifies the collusive CQA spam using Q&A features and the correlations among Q&As. First, we define the Deceptive Pattern of Q&As, based on which the real Q&A groups are extracted. Then, we extract several highly discriminative Q&A features from both individual and group levels, and propose several types of correlations, which correlate the Q&As that are more likely to have the same labels. After uniformly modeling the Q&As, features, and correlations in the Attributed Heterogeneous Information Network (AHIN), a semi-supervised collective classification algorithm is proposed to detect the collusive Q&A spam. Experimental results on a real-life dataset demonstrate that CCQA can accurately detect the collusive CQA spam, and outperform a number of competitive baselines. [ABSTRACT FROM AUTHOR]

Automata and Petri nets are two typical models of discrete event systems. The paper studies the problem of converting a finite automaton into a Petri net that satisfies the specified structural features. More specifically, we identify a labeled Petri net from a nondeterminstic finite automaton such that the reachability graph of the identified net is isomorphic to the given automaton, meaning that the Petri net models exactly the same dynamic system as the automaton. The identified net necessarily satisfies the specified structural requirements, such as the numbers of places and transitions, absence of self-loops, the exact structure of a part of the net, and so on. Since label Petri nets and nondeterminstic finite automata are generalizations of Petri nets and deterministic finite automata, respectively, the proposed approach can identify a larger spectrum of Petri nets and effectively handle the nondeterminstic cases in the given automaton. Four kinds of conditions are first extracted from the given automaton, namely deterministic enabling conditions, nondeterministic enabling conditions (i.e., multiple transitions with the same label are enabled simultaneously at a marking), transition disabling conditions, and marking inequality conditions. Then, an integer linear programming problem is formulated by characterizing these four kinds of conditions using integer linear constraints. A labeled Petri net satisfying the structural requirements is obtained by solving the integer linear programming problem. We also present two examples to show the possible applications of the proposed identification approach. [ABSTRACT FROM AUTHOR]

Various variants have been proposed to improve the search ability and efficiency of Differential Evolution (DE). However, the variants ignore the impact caused by the use of accumulated historical information, resulting in unpromising performance of the search. Furthermore, the global exploration and local exploitation ability of population is affected by the population number (NP), which commonly adapts with little adaptive control or with linear descendent only. To utilize more historical population information, we propose in this paper a novel adaptive population size-based DE (APSDE) by mining historical population information, to balance global exploration and local exploitation ability. APSDE uses historical population information to mine population distribution and extract information to assign parameters to the current population. Moreover, an archive is utilized to store the historical population information, where historical successful parameter information consists of scaling factor (F), crossover rate (CR) and NP for better balancing the search ability of population. To evaluate the performance of APSDE, we conduct experiments both on the 28 benchmark functions of CEC2017, the 10 benchmark functions of CEC2020, and a real-world application - the path planning of unmanned aerial vehicles (UAVs). The experimental results have demonstrated that APSDE achieves the promising performance in both convergence accuracy and convergence speed. [ABSTRACT FROM AUTHOR]

E-commerce order fulfillment service (E-COFS) plays a pivotal role in shaping consumer behavior in online marketplaces. The strategic outsourcing of the service allows e-commerce sellers to prioritize their core business areas, enhance customer satisfaction, and minimize fulfillment costs. However, a critical challenge lies in appraising the potential quality of E-COFS provided by third parties, especially when lacking historical information. To address this, this paper first designs a generalized framework for guiding the construction of the quantitative model for evaluating the potential quality of E-COFS. The proposed framework unfolds in three stages: (1) evaluating potential effectiveness of an E-COFS through quantifying stakeholders' potential satisfaction from the E-COFS plan tailored by its provider, (2) evaluating its potential feasibility by quantifying the potential performance of the E-COFS quality management system (E-COFS-QMS) built by the provider on supporting the plan, and (3) integrating the above two parts to gauge the potential quality of the E-COFS. Building upon this framework, this paper then designs a novel quantitative model. Specifically, this model adopts the linguistic subjective judgment representation method and introduces basic uncertain linguistic information to achieve computing with words. Multiple stakeholders within e-commerce sellers are tasked with articulating their requirements, their preferences and expectations, and consensus reaching process is conducted to obtain the acceptable consensus among these stakeholders. Multiple experts from various domains are tasked with giving their subjective judgements on the performances of E-COFS and E-COFS-QMS, and a method of weighting individual judgments, which respects the reliabilities of individual judgements and the overall similarity in knowledge structures among the experts, is adopted to effectively tap into collective intelligence. Finally, a case study is conducted to validate the validity and feasibility of the proposed quantitative model. [ABSTRACT FROM AUTHOR]