Soft Computing最新文献

The fuzzy concept lattice is one of the effective tools for data mining, and granular reduction is one of its significant research contents. However, little research has been done on granular reduction at different granularities in formal fuzzy contexts (FFCs). Furthermore, the complexity of the composition of the fuzzy concept lattice limits the interest in its research. Therefore, how to simplify the concept lattice structure and how to construct granular reduction methods with granularity have become urgent issues that need to be investigated. To this end, firstly, the concept of an object granule with granularity is defined. Secondly, two reduction algorithms, one based on Boolean reasoning and the other on a graph-theoretic heuristic, are formulated while keeping the structure of this object granule unchanged. Further, to simplify the structure of the fuzzy concept lattice, a partial order relation with parameters is proposed. Finally, the feasibility and effectiveness of our proposed reduction approaches are verified by data experiments.

This study focuses on unspecified dynamic seru scheduling problems with resource constraints (UDSS-R) in seru production system (SPS). A mixed integer linear programming model is formulated to minimize the makespan, which is solved sequentially from both allocation and scheduling perspectives by a strip-packing constructive algorithm (SPCA) with deep reinforcement learning (DRL). The training samples are trained by the DRL model, and the reward values obtained are calculated by SPCA to train the network so that the agent can find a better solution. The output of DRL is the scheduling order of jobs in serus, while the solution of UDSS-R is solved by SPCA. Finally, a set of test instances are generated to conduct computational experiments with different instance scales for the DRL-SPCA, and the results confirm the effectiveness of proposed DRL-SPCA in solving UDSS-R with more outstanding performance in terms of solution quality and efficiency, across three data scales (10 serus × 100 jobs, 20 serus × 250 jobs, and 30 serus × 400 jobs), compared with GA and SAA, the Avg. RPD of DRL-SPCA decreased by 9.93% and 7.56%, 13.36% and 10.72%, and 9.09% and 7.08%, respectively. In addition, the Avg. CPU time was reduced by 29.53% and 27.93%, 57.48% and 57.04%, and 61.73% and 61.76%, respectively.

This study analyzes the conditional gradient method for constrained multiobjective optimization problems, also known as the Frank–Wolfe method. We assume that the objectives are continuously differentiable, and the constraint set is convex and compact. We employ an average-type nonmonotone line search, which takes the average of the recent objective function values. The asymptotic convergence properties without convexity assumptions on the objective functions are established. We prove that every limit point of the sequence of iterates that is obtained by the proposed method is a Pareto critical point. An iteration-complexity bound is provided regardless of the convexity assumption on the objective functions. The effectiveness of the suggested approach is demonstrated by applying it to several benchmark test problems. In addition, the efficiency of the proposed algorithm in generating approximations of the entire Pareto front is compared to the existing Hager–Zhang conjugate gradient method, the steepest descent method, the monotone conditional gradient method, and a nonmonotone conditional gradient method. In finding empirical comparison, we utilize two commonly used performance matrices—inverted generational distance and hypervolume indicators.

Chest X-ray imaging is a critical diagnostic tool for identifying pulmonary diseases. However, manual interpretation of these images is time-consuming and error-prone. Automated systems utilizing convolutional neural networks (CNNs) have shown promise in improving the accuracy and efficiency of chest X-ray image classification. While previous work has mainly focused on using feature maps from the final convolution layer, there is a need to explore the benefits of leveraging additional layers for improved disease classification. Extracting robust features from limited medical image datasets remains a critical challenge. In this paper, we propose a novel deep learning-based multilayer multimodal fusion model that emphasizes extracting features from different layers and fusing them. Our disease detection model considers the discriminatory information captured by each layer. Furthermore, we propose the fusion of different-sized feature maps (FDSFM) module to effectively merge feature maps from diverse layers. The proposed model achieves a significantly higher accuracy of 97.21% and 99.60% for both three-class and two-class classifications, respectively. The proposed multilayer multimodal fusion model, along with the FDSFM module, holds promise for accurate disease classification and can also be extended to other disease classifications in chest X-ray images.

In this article, the concept of fuzzy (m, n)-quasi-ideals in ordered semigroups is developed and discussed in various ways. In addition, we present the concepts of fuzzy (m, 0)-ideals and fuzzy (0, n)-ideals in ordered semigroups and investigate some of their associated properties. Furthermore, the (m, n)-regular ordered semigroups are studied in terms of fuzzy (m, n)-quasi-ideals, fuzzy (m, 0)-ideals, and fuzzy (0, n)-ideals.

Several constrained optimization problems have been adequately solved over the years thanks to the advances in the area of metaheuristics. Nevertheless, the question as to which search logic performs better on constrained optimization often arises. In this paper, we present Dual Search Optimization (DSO), a co-evolutionary algorithm that includes an adaptive penalty function to handle constrained problems. Compared to other self-adaptive metaheuristics, one of the main advantages of DSO is that it is able auto-construct its own perturbation logics, i.e., the ways solutions are modified to create new ones during the optimization process. This is accomplished by co-evolving the solutions (encoded as vectors of integer/real values) and perturbation strategies (encoded as Genetic Programming trees), in order to adapt the search to the problem. In addition to that, the adaptive penalty function allows the algorithm to handle constraints very effectively, yet with a minor additional algorithmic overhead. We compare DSO with several algorithms from the state-of-the-art on two sets of problems, namely: (1) seven well-known constrained engineering design problems and (2) the CEC 2017 benchmark for constrained optimization. Our results show that DSO can achieve state-of-the-art performances, being capable to automatically adjust its behavior to the problem at hand.

Under the wave of the digital era, the retail industry is facing unprecedented fierce competition and a rapidly changing market environment. In this context, developing smart and efficient pricing strategies has become a top priority in the industry. Faced with this challenge, traditional pricing methods are inadequate due to their slow response, insufficient adaptability to instant changes in the market, and over-reliance on historical data and human experience. In response to this urgent need, this study aims to design an intelligent pricing model rooted in deep learning to enhance the vitality and competitiveness of the retail industry. The emerging solution adopted in this article combines Temporal Fusion Transformer (TFT), Ensemble of Simplified RNNs (ES-RNN), and dynamic attention mechanisms, aiming to accurately capture and analyze complex time series data through these advanced technologies. TFT processes multivariate and multi-level data, ES-RNN technology integrates multiple simple versions of recurrent neural networks to enhance predictive power, and the dynamic attention mechanism allows the model to dynamically weight the importance of different points in the time series, thereby improving the effectiveness of feature extraction. Test experimental results on four different data sets show that our models all show excellent performance, and the accuracy of predicted product sales far exceeds traditional models. In addition, with its ability to dynamically adjust pricing, the model demonstrates excellent stability and adaptability amid market fluctuations. This research not only promotes the intelligent transformation of retail pricing strategies, but also provides a more strategic tool for enterprises to compete for market share.

Renyi entropy-based thresholding is a popular image segmentation method. In this work, to improve the performance of the Renyi entropy thresholding method, an efficient adaptive multilevel Renyi entropy thresholding method based on the energy curve with dynamic programming (DP + ARET) is presented. First, the histogram is substituted by the energy curve in the Renyi entropy thresholding to take advantage of the spatial context information of pixels. Second, an adaptive entropy index selection strategy is proposed based on the image histogram. Finally, to decrease the computation complexity of the multilevel Renyi entropy thresholding, an efficient solution is calculated by the dynamic programming technique. The proposed DP + ARET method can obtain the global optimal thresholds with the time complexity linear in the number of the thresholds. The comparative experiments between the proposed method with the histogram-based method verified the effectiveness of the energy curve. The segmentation results on the COVID-19 Computed Tomography (CT) images with the same objective function by the proposed DP + ARET and swarm intelligence optimization methods testify that the DP + ARET can quickly obtain the global optimal thresholds. Finally, the performance of the DP + ARET method is compared with several image segmentation methods quantitatively and qualitatively, the average segmented accuracy (SA) is improved by 7% than the comparative methods. The proposed DP + ARET method can be used to fast segment the images with no other prior knowledge.

The saccade is one of the eye movements that resulted in the creation of the saccadic model. This work is grounded in the basic principles of the saccadic system, which are burst neurons and a resettable integrator model. Considering the possibility of strengthening the saccadic model based on its fundamental model, we introduce a replacement function for use in the burster equation that explains the preservation of the on response’s form and also considers the off response. The new model is a two-dimensional map containing slow and fast variables with a new burster function, which solves the lack of differentiability of the primary function at the equilibrium point. By applying time series approaches and phase portraits, the mechanisms underlying the generation of spikes and spike bursts in the behavior of the new model are revealed. The present research’s other main focus is to determine the geometry of the slow manifold for the newly developed system. Specifically, we examine the dynamics around an equilibrium point and the geometry of a slow manifold by using Fenichel’s theorem. In addition, we use the center manifold theory to describe some dynamical characteristics of the center manifold that the slow manifold matches. Finally, this study aims to figure out the effects of geometric singular perturbations on this fast-slow burster equation, which finds dynamical behaviors such as being uniformly asymptotically stable and locally attractive.

In recent years, the method of using graph neural networks (GNN) to learn users’ social influence has been widely applied to social recommendation and has shown effectiveness, but several important challenges have not been well addressed: (i) Most work fails to consider the user interests (historical user-item interactions) when first building user-user social relationships, which can make it difficult to capture accurate user embedding and thus prevent the model from better exploring the users’ social influence; (ii) Most of the current methods do not build social neighbors (with the same item-user interaction) that belong to the item and do not aggregate information from the perspective of social neighbors, which makes it possible for the item to lose a lot of details when expressing the user’s interest factors. Therefore, to address the above challenges, we propose Exploring Implicit Influence for Social Recommendation Based on GNN (EIIGNN). First, we construct the initial user embedding with user-item interaction information and use the implicit modeling module in user modeling to explore the implicit influence of interest factors on users. In addition, EIIGNN models the social graph structure of item (an item-item graph) so that item can aggregate information from the perspective of their social neighbors, which helps the model learn a more accurate representation of the item. Finally, extensive experimental results on two real-world datasets clearly demonstrate the effectiveness of EIIGNN.