International Journal of Machine Learning and Cybernetics最新文献

The tourism industry increasingly employs sensors and machine learning for tasks such as demand prediction and mobility forecasting. However, some challenges in data collection remain, especially with information privacy and resource management. We propose a vehicle classification model based on License Plate Recognition (LPR) sensor data, incorporating contextual datasets not explored in the existing literature to predict the number of nights a vehicle will stay in a mountain tourist area. We also study the importance of each dataset in the results. Our analysis utilizes data from four LPR cameras spanning 17 months. We compare different classification models optimized through ensemble techniques. Additionally, an ablation study assesses the impact of each dataset, with variables categorized by expert knowledge into seasonal, socio-economic or visit-related. Optimal dataset selection demonstrates a 22.2% reduction in processing time and an 80% decrease in the number of variables, with only a slight decrease of 0.01 in the Area Under the Curve (AUC) compared to using all available variables. This research provides information to develop tourism prediction models, guiding which datasets and calculated variables are the most important while balancing the processing time and AUC.

This research introduces a new method to tackle the issue of exchanging cryptographic keys in the Industrial Internet of Things (IIoT). This study focuses on the inefficiency and lengthy evaluation procedures of conventional cryptographic key exchange algorithms, which are not appropriate for the rapid and constantly changing IIoT device environment. In the solution domain, the proposed approach uses synchronization of neural networks with vector valued and Recurrent Neural Networks (RNNs), merging drive-response mechanisms to enhance speed and efficiency in crucial operations. The research examines the influence of postponements on the generating arbitrary inputs and coordination challenges in RNNs that incorporate drive-response mechanisms for synchronized input vector creation. This article explains an elementary evaluation of coordination in Artificial Neural Networks (ANNs) by utilizing an RNN framework to structure ANNs for sharing session keys. The study provides multiple contributions: (1) employing the polynomial coordination technique to generate coordinated inputs for the ANN synchronization process using RNNs, (2) using Lyapunov formulas and inequality assessment methods to identify required control parameters and time-varying conditions for achieving synchronization in the drive-response systems proposed with polynomial and non-polynomial functions, (3) demonstrating the connection between polynomial and non-polynomial synchronization with numerical illustrations, and (4) designing symmetric layouts of ANNs to create a session keys in the IIoT network. The suggested technique outperforms existing methods in the literature by offering a quicker, more dependable solution for cryptographic key exchange, paving the way for improved and secure industrial applications. This new method not only fixes current inefficiencies but also paves the way for future improvements in secure communication in the IIoT environment.

Emergency supply chain management has recently drawn growing attention of managers and researchers with frequent appearance of pandemics, disasters and safety accidents. Previous studies proposed methods for supplier selection and order allocation, while they cannot satisfy the demand for emergency supplies as emergency events bring many uncertainties and risks in supply chain disruption. To guarantee the efficiency in emergency supplies procurement, this work aims at putting forward a two-stage approach for emergency supplier selection and order allocation by use of three-way decision and fuzzy multi-objective optimization. Firstly, by considering the perceived utilities and perceived losses of purchasing process simultaneously, a prospect profit-based three-way decision model is established. Next, the prospect multi-attribute three-way decision model for emergency supplier selection is proposed, constructing the calculation approaches of thresholds, conditional probabilities as well as decision rules. Thirdly, inspired by perceived utilities and perceived losses of supplies purchasing, the utility-based objective function and loss-based objective function are introduced to multi-objective optimization model for order allocation. Finally, a real case of government emergency supplies procurement is discussed to show the applicability and effectiveness of the proposed approach. The final results of the proposed methodology show that it can effectively manage data with uncertainty, determine the qualified suppliers as well as alternative suppliers simultaneously to prevent emergency supply chain disruption, and provide satisfactory solutions for order allocation by introducing different combinations of objective functions according to decision makers’ preference.

The constrained resources of Internet of Things (IoT) devices make them susceptible to Distributed Denial-of-Service (DDoS) attacks that disrupt service availability by overwhelming systems. Thus, effective intrusion detection is critical to ensuring uninterrupted IoT activities. This research presents a scalable system that combines machine and deep learning models with optimized data processing to secure IoT devices against DDoS attacks. A real-world 5G-IoT network simulation dataset was used to evaluate performance. Robust feature selection identified the 10 most informative features from the high-dimensional data. These features were used to train eight classifiers, namely: k-Nearest Neighbors (KNN), Naive Bayes (NB), Decision Tree (DT), Random Forest (RF), Multi-Layer Perceptron (MLP), Convolutional Neural Network (CNN), Long-Short-Term Memory (LSTM) and hybrid CNN-LSTM models for DDoS attack detection. Experiments demonstrated 99.99% and 99.98% accuracy for multiclass and binary classification using the proposed hybrid CNN-LSTM model. Crucially, time- and space-complexity analysis validates real-world feasibility. Unlike prior works, this system optimally balances accuracy, efficiency, and adaptability through a precisely engineered model architecture, outperforming existing models. In general, this accurate, efficient, and adaptable system addresses critical IoT security challenges, improving cyber resilience in smart cities and autonomous vehicles.

As a fundamental image characteristic, edge features encapsulate a wealth of information, serving as a crucial foundation in image segmentation networks for accurately delineating and partitioning object edges. Convolutional neural networks (CNNs) have gained prominence recently, finding extensive utility in edge detection. Previous methods primarily emphasized edge prediction accuracy, ignoring edge refinement. In this work, we introduce a novel encoder-decoder architecture that effectively harnesses hierarchical features. By extending the decoder horizontally, we progressively enhance resolution to preserve intricate details from the original image, thereby producing sharp edges. Additionally, we propose a novel loss function named the Pixel-Patch Combination Loss (P²CL), which employs distinct detection strategies in edge and non-edge regions to bolster network accuracy and yield crisp edges. Furthermore, considering the practicality of the algorithm, our method strikes a fine balance between accuracy and model size. It delivers precise and sharp edges while ensuring efficient model operation, thereby laying a robust foundation for advancements deployed on mobile devices or embedded systems. Our method was evaluated on three publicly available datasets, including BSDS500, Multicue, and BIPED. The experimental results show the superiority of our approach, achieving a competitive ODS F-score of 0.832 on the BSDS500 benchmark and significantly enhancing edge detection accuracy.

({mathbf{I}}_{{mathbf{SDE}}^{+}}) is proven to be one of the leading scalable indicator for evolutionary multi and many-objective optimization. However, it fails to segregate members of a given population beyond the first front as a large number of solutions in the population have identical ({mathbf{I}}_{{mathbf{SDE}}^{+}}) values. This mainly affects the performance of the algorithm when handling optimization problems with lower objectives. Consequently, we hypothesize that the overall performance of the algorithm can be further improved by introducing a categorization mechanism similar to the categorization of Pareto Fronts (PFs) in dominance-based methods. Therefore, in this work, we propose a Multi-Indicator-Based Multi-Objective Evolutionary Algorithm (MI-MOEA) which categorizes all the solutions into different fronts. Specifically, the indicators are based on the popular ({mathbf{I}}_{{mathbf{SDE}}^{+}}) indicator and make use of the minimum and median distance values among the different distances when the solutions with better Sum of Objectives (SOB) are projected. The use of these two ({mathbf{I}}_{{mathbf{SDE}}^{+}})-based indicator values features an efficient balance of exploration and exploitation. To evaluate the performance of the proposed MI-MOEA, Neural Architecture Search (NAS) which involves the design of appropriate architectures suitable for specific applications is employed. From an optimization perspective, NAS involves multiple conflicting objectives that needs to be simultaneously optimized. In this paper, we consider a recently proposed multi-objective NAS benchmark and favorably evaluate the performance of MI-MOEA compared to other state-of-the-art MOEAs.

To improve the generalizability and robustness of stochastic configuration networks (SCNs), this paper proposes a robust modeling method based on information superposition and mixture correntropy. First, the mapping information of the (sigmoid) activation function and its derivative function is superimposed, and the hidden layer parameters are randomly assigned through a supervisory mechanism to improve the diversity of the hidden layer mapping. Second, mixture correntropy is used to construct a robust loss function, and different Gaussian kernels are used to measure the contribution of training samples to suppress the negative impact of data noise on the accuracy of the model. Finally, the performance of the proposed modeling method is tested on functional approximation, four benchmark datasets, and historical data from the municipal solid waste incineration process. The experimental results show that the modeling method proposed in this paper has advantages in terms of generalizability and robustness.

Label enhancement (LE) refers to the process of recovering label distributions from logical labels for less ambiguity. Current LE techniques concentrate on learning each instance individually, which ignores the instance correlation. In this paper, we propose to learn a cluster-wise label distribution (CWLD) shared by all instances of the cluster to explore the instance correlation. The softmax-normalized sum of the CWLD and the logical label vector yields the label distribution. CWLD is learned in an iterative manner. Following instance clustering, the label distributions of all instances in each cluster are averaged. The asymmetric label correlation is then mined using heat conduction. This process is repeated until the label distribution has reached a point of convergence. Experiments were undertaken on thirteen real-world datasets compared with six state-of-the-art algorithms. Results demonstrate the effectiveness and superiority of our proposed method.

The mean–variance (MV) model provides numerous optimal portfolios for managing a firm's asset portfolio. Portfolio decisions in large corporations involve many interest groups, such as shareholders, bondholders, and employees, and require the assistance of large experts. However, experts from different departments with different cognitive levels and interests can differ or even conflict in their assessments of portfolios. To guarantee their interests, some experts may exhibit non-cooperative behavior, thus reducing the efficiency of reaching a consensus. To tackle this issue, the research aims to develop a large-scale group interactive portfolio optimization method that incorporates non-cooperative behaviors and leverages social network analysis (SN-LSGDM-NC-PO). First, various consensus feedback strategies based on minimum adjustment are formulated to provide advice during the negotiation process according to the global and local levels. Then, considering the acceptance of advice and the effect of expert adjustment on consensus, a new measure of non-cooperative behavior is designed. Non-cooperative behavior by experts can affect trust relations in a social network. Therefore, trust reward and penalty mechanisms, preference penalty mechanisms, and an exit mechanism are developed to manage different types of non-cooperative behavior. Experimental and comparison results demonstrate that the proposed SN-LSGDM-NC-PO algorithm can effectively manage the non-cooperative behaviors and reduce interaction consensus costs.

Single object tracking is a vital task of many applications in critical fields. However, it is still considered one of the most challenging vision tasks. In recent years, computer vision, especially object tracking, witnessed the introduction or adoption of many novel techniques, setting new fronts for performance. In this survey, we visit some of the cutting-edge techniques in vision, such as Sequence Models, Generative Models, Self-supervised Learning, Unsupervised Learning, Reinforcement Learning, Meta-Learning, Continual Learning, and Domain Adaptation, focusing on their application in single object tracking. We propose a novel categorization of single object tracking methods based on novel techniques and trends. Also, we conduct a comparative analysis of the performance reported by the methods presented on popular tracking benchmarks. Moreover, we analyze the pros and cons of the presented approaches and present a guide for non-traditional techniques in single object tracking. Finally, we suggest potential avenues for future research in single-object tracking.