Pub Date : 2026-01-14DOI: 10.1016/j.asoc.2026.114661
Michal Burda
In pattern mining from tabular data using fuzzy logic, a common task involves computing triangular norms (t-norms) to represent conjunctions of fuzzy predicates and summing the resulting truth values to evaluate rule support or other pattern quality measures. Building on previous work, this paper presents an approach that packs multiple fuzzy truth values into a single integer and performs t-norm computations directly on this compact representation. By using 4-, 8-, or 16-bit precision, the method substantially reduces memory consumption and improves computational efficiency. For example, with 8-bit precision—offering two decimal places of accuracy—it requires only one-quarter of the memory and achieves 3–16 speedup compared to conventional floating-point-based method of computation. The proposed method is also compared with a traditional computation approach optimized using advanced Single-Instruction/Multiple-Data (SIMD) CPU operations, demonstrating its superior performance on modern architectures.
{"title":"Accelerating pattern mining on fuzzy data by packing truth values into blocks of bits","authors":"Michal Burda","doi":"10.1016/j.asoc.2026.114661","DOIUrl":"10.1016/j.asoc.2026.114661","url":null,"abstract":"<div><div>In pattern mining from tabular data using fuzzy logic, a common task involves computing triangular norms (t-norms) to represent conjunctions of fuzzy predicates and summing the resulting truth values to evaluate rule support or other pattern quality measures. Building on previous work, this paper presents an approach that packs multiple fuzzy truth values into a single integer and performs t-norm computations directly on this compact representation. By using 4-, 8-, or 16-bit precision, the method substantially reduces memory consumption and improves computational efficiency. For example, with 8-bit precision—offering two decimal places of accuracy—it requires only one-quarter of the memory and achieves 3–16<span><math><mo>×</mo></math></span> speedup compared to conventional floating-point-based method of computation. The proposed method is also compared with a traditional computation approach optimized using advanced Single-Instruction/Multiple-Data (SIMD) CPU operations, demonstrating its superior performance on modern architectures.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"191 ","pages":"Article 114661"},"PeriodicalIF":6.6,"publicationDate":"2026-01-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146039728","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-01-14DOI: 10.1016/j.asoc.2026.114604
Zitang Zhang, Qian Sun, Yujie Huang, Yibing Li
With the advancement of unmanned systems technology, unmanned aerial vehicle (UAV) swarms composed of miniaturized, heterogeneous, and intelligent platforms have emerged as a new operational paradigm in military operations. However, efficient heterogeneous UAV swarm payload configuration (HUPC) remains a significant challenge due to limited single-platform capabilities, which is further exacerbated by the increasing number of platforms, strong coupling among diverse resource types, and heterogeneous mission requirements. To address this issue, this paper proposes a payload configuration algorithm tailored to the operational characteristics of UAV swarms. The HUPC problem is formulated as a bi-variable integer nonlinear programming model, with the objective of minimizing overall configuration cost while satisfying multiple operational constraints. To solve the above model, a heuristic initialization strategy is developed based on multi-attribute encoding and task-driven prioritization, combined with a parallel evolutionary and variable neighborhood search (VNS) approach under the genetic algorithm’s (GA) framework. The algorithm leverages accumulated historical experience during the optimization process to efficiently derive payload configuration schemes that meet mission requirements. Simulation results demonstrate that, compared with existing approaches, the proposed method reduces the average configuration cost by approximately 10% in small-scale scenarios and about 7% in medium- and large-scale scenarios, while maintaining stable performance under multiple operational constraints. This demonstrates that the proposed method ensures the feasibility and rationality of payload configuration schemes across tasks of varying scales.
{"title":"A heuristic payload configuration method for UAV swarm based on hybrid genetic algorithm and variable neighborhood search","authors":"Zitang Zhang, Qian Sun, Yujie Huang, Yibing Li","doi":"10.1016/j.asoc.2026.114604","DOIUrl":"10.1016/j.asoc.2026.114604","url":null,"abstract":"<div><div>With the advancement of unmanned systems technology, unmanned aerial vehicle (UAV) swarms composed of miniaturized, heterogeneous, and intelligent platforms have emerged as a new operational paradigm in military operations. However, efficient heterogeneous UAV swarm payload configuration (HUPC) remains a significant challenge due to limited single-platform capabilities, which is further exacerbated by the increasing number of platforms, strong coupling among diverse resource types, and heterogeneous mission requirements. To address this issue, this paper proposes a payload configuration algorithm tailored to the operational characteristics of UAV swarms. The HUPC problem is formulated as a bi-variable integer nonlinear programming model, with the objective of minimizing overall configuration cost while satisfying multiple operational constraints. To solve the above model, a heuristic initialization strategy is developed based on multi-attribute encoding and task-driven prioritization, combined with a parallel evolutionary and variable neighborhood search (VNS) approach under the genetic algorithm’s (GA) framework. The algorithm leverages accumulated historical experience during the optimization process to efficiently derive payload configuration schemes that meet mission requirements. Simulation results demonstrate that, compared with existing approaches, the proposed method reduces the average configuration cost by approximately 10% in small-scale scenarios and about 7% in medium- and large-scale scenarios, while maintaining stable performance under multiple operational constraints. This demonstrates that the proposed method ensures the feasibility and rationality of payload configuration schemes across tasks of varying scales.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"190 ","pages":"Article 114604"},"PeriodicalIF":6.6,"publicationDate":"2026-01-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145979794","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-01-14DOI: 10.1016/j.asoc.2026.114656
Xuan Zhang , Boyu Hu , Xusheng Sun , Jingling Ma , Gang Wang , Tingting (Rachel) Chung
Internal control material weaknesses (ICMW) are often early warnings of possible financial misstatements or fraud that can lead to financial distress. Therefore, accurately predicting ICMW is crucial to mitigating greater losses. Recent studies have shown that multi-modal data holds significant promise for predicting ICMW in listed companies. However, the complementary effects of multi-modal data remain underexplored. This limits the model’s ability to fully capture the ICMW clues. Furthermore, existing studies primarily treat companies as independent entities. They overlooked the inter-company relationships that may influence the final prediction results. To address these limitations, this study proposes a Multi-level Attentive Graph Fusion with Cross-modal Complementary Learning (MAGF-CCL) method for ICMW prediction. Specifically, first, the instance-level graphs are constructed using k-nearest neighbors (KNN) algorithm. Graph Convolutional Network (GCN) is then employed to learn inter-company relationships in graphs. Second, a Multi-modal Complementary Learning (MCL) module is designed to explore the multi-modal complementarity, hence fully capturing ICMW clues. Third, to integrate multi-modalities effectively, the numerical and textual graphs are fused using Modality-level Fusion Mechanism (MFM) and Structure-level Fusion Mechanism (SFM). These fusion modules combine the multi-modal data and structural relationships, respectively. Finally, the fused graph is subsequently fed into a GCN to facilitate cross-modal information propagation and enhance ICMW prediction. Experimental results on a real-world dataset demonstrate that the proposed MAGF-CCL method outperforms state-of-the-art (SOTA) methods in predicting ICMW. The AUC value of MAGF-CCL achieved 91.04 %, outperforming existing SOTA method by nearly 3 %. This study also visualized the inter-company relationships and attention maps of MFM module, thereby providing relevant decision support for stakeholders.
{"title":"MAGF-CCL: Multi-level attentive graph fusion with cross-modal complementary learning for internal control material weaknesses prediction","authors":"Xuan Zhang , Boyu Hu , Xusheng Sun , Jingling Ma , Gang Wang , Tingting (Rachel) Chung","doi":"10.1016/j.asoc.2026.114656","DOIUrl":"10.1016/j.asoc.2026.114656","url":null,"abstract":"<div><div>Internal control material weaknesses (ICMW) are often early warnings of possible financial misstatements or fraud that can lead to financial distress. Therefore, accurately predicting ICMW is crucial to mitigating greater losses. Recent studies have shown that multi-modal data holds significant promise for predicting ICMW in listed companies. However, the complementary effects of multi-modal data remain underexplored. This limits the model’s ability to fully capture the ICMW clues. Furthermore, existing studies primarily treat companies as independent entities. They overlooked the inter-company relationships that may influence the final prediction results. To address these limitations, this study proposes a Multi-level Attentive Graph Fusion with Cross-modal Complementary Learning (MAGF-CCL) method for ICMW prediction. Specifically, first, the instance-level graphs are constructed using k-nearest neighbors (KNN) algorithm. Graph Convolutional Network (GCN) is then employed to learn inter-company relationships in graphs. Second, a Multi-modal Complementary Learning (MCL) module is designed to explore the multi-modal complementarity, hence fully capturing ICMW clues. Third, to integrate multi-modalities effectively, the numerical and textual graphs are fused using Modality-level Fusion Mechanism (MFM) and Structure-level Fusion Mechanism (SFM). These fusion modules combine the multi-modal data and structural relationships, respectively. Finally, the fused graph is subsequently fed into a GCN to facilitate cross-modal information propagation and enhance ICMW prediction. Experimental results on a real-world dataset demonstrate that the proposed MAGF-CCL method outperforms state-of-the-art (SOTA) methods in predicting ICMW. The AUC value of MAGF-CCL achieved 91.04 %, outperforming existing SOTA method by nearly 3 %. This study also visualized the inter-company relationships and attention maps of MFM module, thereby providing relevant decision support for stakeholders.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"191 ","pages":"Article 114656"},"PeriodicalIF":6.6,"publicationDate":"2026-01-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145986733","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-01-14DOI: 10.1016/j.asoc.2026.114621
Long Shi , Junyu Chen , Lei Cao , Jun Wang , Jinghua Tan , Badong Chen
Recently, multi-view Graph Neural Networks (GNNs) have garnered increasing interest. However, three critical research aspects still remain challenging: 1) capturing underlying correlation information between views, 2) extracting intrinsic graph structure features, and 3) aggregating graph information from different views. To address these challenges, we propose a novel multi-view graph neural network framework. Specifically, we capture the local correlation between the views in the kernel feature space. By stacking the mapped graph matrices into a tensor, tensor decomposition is then performed to extract the global correlation among different graphs, which enhances both the adjacency and feature matrices. To explore the inherent graph structure features, we design an unsupervised scheme for filtering out low-relevance neighbors. This is achieved by initially constructing a score matrix based on similarity measures to evaluate the neighbor importance, and then designing a node-filtering strategy to balance important neighbors and fruitful edges. Finally, we design an augmented cross-aggregation module to enable in-depth intra-aggregation and inter-aggregation. Experimental results on real-world datasets show that our method outperforms several advanced graph neural network methods. The code will soon be released in a preprint version.
{"title":"Multi-view graph neural networks by augmented aggregation","authors":"Long Shi , Junyu Chen , Lei Cao , Jun Wang , Jinghua Tan , Badong Chen","doi":"10.1016/j.asoc.2026.114621","DOIUrl":"10.1016/j.asoc.2026.114621","url":null,"abstract":"<div><div>Recently, multi-view Graph Neural Networks (GNNs) have garnered increasing interest. However, three critical research aspects still remain challenging: 1) capturing underlying correlation information between views, 2) extracting intrinsic graph structure features, and 3) aggregating graph information from different views. To address these challenges, we propose a novel multi-view graph neural network framework. Specifically, we capture the local correlation between the views in the kernel feature space. By stacking the mapped graph matrices into a tensor, tensor decomposition is then performed to extract the global correlation among different graphs, which enhances both the adjacency and feature matrices. To explore the inherent graph structure features, we design an unsupervised scheme for filtering out low-relevance neighbors. This is achieved by initially constructing a score matrix based on similarity measures to evaluate the neighbor importance, and then designing a node-filtering strategy to balance important neighbors and fruitful edges. Finally, we design an augmented cross-aggregation module to enable in-depth intra-aggregation and inter-aggregation. Experimental results on real-world datasets show that our method outperforms several advanced graph neural network methods. The code will soon be released in a preprint version.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"191 ","pages":"Article 114621"},"PeriodicalIF":6.6,"publicationDate":"2026-01-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146039723","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-01-14DOI: 10.1016/j.asoc.2026.114662
Gurmeet Saini, Shimpi Singh Jadon
Evolutionary algorithms often suffer from search inefficiency due to their inability to systematically reuse historical search patterns, leading to redundant exploration and premature stagnation. Addressing this limitation, we propose KLABC-RL, a novel framework that synergizes Reinforcement Learning (RL) with Knowledge Learning Evolutionary Computation (KLEC) within the Artificial Bee Colony (ABC) paradigm. Unlike conventional hybrids that enforce static knowledge transfer, KLABC-RL employs a Q-learning-based adaptive agent to dynamically govern the search process. This agent intelligently toggles between an Artificial Neural Network (ANN)-driven Knowledge Learning Model (KLM) for exploitation and standard ABC operators for exploration, thereby effectively preventing negative knowledge transfer. To further mitigate stagnation, a Hilbert space-based perturbation strategy is integrated into the scout phase, enhancing population diversity. Comprehensive evaluations on 23 classical benchmark functions, the IEEE CEC 2019 suite, and complex real-world engineering problems, specifically planar kinematic arm control and photovoltaic (PV) parameter extraction demonstrate the superiority of KLABC-RL. Comparative analysis against seven state-of-the-art algorithms and 4 hybrid variants of ABC reveals that KLABC-RL achieves significantly faster convergence and higher solution accuracy. Rigorous statistical validation, including Wilcoxon Rank-Sum, Friedman, and ANOVA tests, confirms the robustness and efficacy of the proposed framework in advancing intelligent evolutionary search.
{"title":"A neural knowledge learning-driven artificial bee colony algorithm with reinforcement adaptation for global optimization","authors":"Gurmeet Saini, Shimpi Singh Jadon","doi":"10.1016/j.asoc.2026.114662","DOIUrl":"10.1016/j.asoc.2026.114662","url":null,"abstract":"<div><div>Evolutionary algorithms often suffer from search inefficiency due to their inability to systematically reuse historical search patterns, leading to redundant exploration and premature stagnation. Addressing this limitation, we propose KLABC-RL, a novel framework that synergizes Reinforcement Learning (RL) with Knowledge Learning Evolutionary Computation (KLEC) within the Artificial Bee Colony (ABC) paradigm. Unlike conventional hybrids that enforce static knowledge transfer, KLABC-RL employs a Q-learning-based adaptive agent to dynamically govern the search process. This agent intelligently toggles between an Artificial Neural Network (ANN)-driven Knowledge Learning Model (KLM) for exploitation and standard ABC operators for exploration, thereby effectively preventing negative knowledge transfer. To further mitigate stagnation, a Hilbert space-based perturbation strategy is integrated into the scout phase, enhancing population diversity. Comprehensive evaluations on 23 classical benchmark functions, the IEEE CEC 2019 suite, and complex real-world engineering problems, specifically planar kinematic arm control and photovoltaic (PV) parameter extraction demonstrate the superiority of KLABC-RL. Comparative analysis against seven state-of-the-art algorithms and 4 hybrid variants of ABC reveals that KLABC-RL achieves significantly faster convergence and higher solution accuracy. Rigorous statistical validation, including Wilcoxon Rank-Sum, Friedman, and ANOVA tests, confirms the robustness and efficacy of the proposed framework in advancing intelligent evolutionary search.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"191 ","pages":"Article 114662"},"PeriodicalIF":6.6,"publicationDate":"2026-01-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146039857","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
High machining accuracy is crucial in CNC turning when manufacturing workpieces. Techniques such as thermal compensation and tool wear prediction reduce errors but rely heavily on accurate machining signals. However, variations in operations, turning tools, machining setting parameters, and noise complicate identifying the actual machining time, posing significant challenges for subsequent tasks. This study proposes a novel framework for identifying machining time intervals and lengths based on the CNC programming language (G-code) and multiple sensor signals. The proposed auto-Seg approach integrates G-code parsing with synchronized data acquisition and signal segmentation, enabling automatic and precise identification of machining states to enhance temporal alignment and enable precise feature extraction without manual intervention. Begin by analyzing G-code to calculate theoretical machining time, followed by utilizing motor signals to identify the actual start and end points of machining. Then, map those vibration signals to extract the actual machining segments. The segmented vibration data is used to pre-train a Convolutional Neural Network (CNN), enabling the model to identify cutting signals and verify their alignment with G-code-defined periods. To validate the proposed auto-Seg approach, various workpieces were tested in different factories. The results showed that the auto-Seg approach accurately identified cutting segments and their corresponding machining durations. It not only demonstrates the effectiveness of the proposed signal synchronization and segmentation framework but also reliably enhances data analytics, monitoring, and diagnostics in CNC machining, using lightweight models suitable for edge deployment and real-world applications.
{"title":"Auto-seg: An automated G-code interpreter and 1DCNN-based framework for signal segmentation and synchronization in CNC machining","authors":"Che-Wei Chou , Hwai-Jung Hsu , Kai-Chun Huang , Yu-Chieh Chen","doi":"10.1016/j.asoc.2026.114644","DOIUrl":"10.1016/j.asoc.2026.114644","url":null,"abstract":"<div><div>High machining accuracy is crucial in CNC turning when manufacturing workpieces. Techniques such as thermal compensation and tool wear prediction reduce errors but rely heavily on accurate machining signals. However, variations in operations, turning tools, machining setting parameters, and noise complicate identifying the actual machining time, posing significant challenges for subsequent tasks. This study proposes a novel framework for identifying machining time intervals and lengths based on the CNC programming language (G-code) and multiple sensor signals. The proposed auto-Seg approach integrates G-code parsing with synchronized data acquisition and signal segmentation, enabling automatic and precise identification of machining states to enhance temporal alignment and enable precise feature extraction without manual intervention. Begin by analyzing G-code to calculate theoretical machining time, followed by utilizing motor signals to identify the actual start and end points of machining. Then, map those vibration signals to extract the actual machining segments. The segmented vibration data is used to pre-train a Convolutional Neural Network (CNN), enabling the model to identify cutting signals and verify their alignment with G-code-defined periods. To validate the proposed auto-Seg approach, various workpieces were tested in different factories. The results showed that the auto-Seg approach accurately identified cutting segments and their corresponding machining durations. It not only demonstrates the effectiveness of the proposed signal synchronization and segmentation framework but also reliably enhances data analytics, monitoring, and diagnostics in CNC machining, using lightweight models suitable for edge deployment and real-world applications.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"191 ","pages":"Article 114644"},"PeriodicalIF":6.6,"publicationDate":"2026-01-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146039882","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-01-14DOI: 10.1016/j.asoc.2026.114660
Zhuojun Han , Yitian Xu
Real-world multi-view datasets are often large and incomplete, driving anchor-based multi-view clustering (MVC) to be extended toward incomplete multi-view clustering (IMVC). Among anchor-based approaches, predefined-anchor methods are attractive due to their high efficiency without iterative anchor refinement. However, when applied to incomplete views, they still face two major challenges: unstable anchor selection and limited utilization of high-order information. These limitations degrade the quality of embedding features and affect clustering performance. To address these challenges, we propose IMVC-TPAL (Efficient Incomplete Multi-View Tensor Clustering through Predefined Anchor Learning) , which begins with a customized anchor selection strategy that reduces randomness and mitigates the impact of missing views, and further incorporates adaptive anchor graph completion directly into the embedding learning process. Additionally, a tensor-based low-frequency approximation operator is employed to explore intra-view similarity, resulting in smooth and discriminative embedding features. In experiments conducted on five datasets under three missing-view ratios, IMVC-TPAL achieves the best performance on 73.3% of all evaluation metrics and ranks second on the remaining ones, demonstrating its effectiveness. These results confirm that our method successfully integrates predefined-anchor learning with the incomplete multi-view setting, providing a reliable and scalable solution for IMVC.
{"title":"Efficient incomplete multi-view tensor clustering through predefined anchor learning","authors":"Zhuojun Han , Yitian Xu","doi":"10.1016/j.asoc.2026.114660","DOIUrl":"10.1016/j.asoc.2026.114660","url":null,"abstract":"<div><div>Real-world multi-view datasets are often large and incomplete, driving anchor-based multi-view clustering (MVC) to be extended toward incomplete multi-view clustering (IMVC). Among anchor-based approaches, predefined-anchor methods are attractive due to their high efficiency without iterative anchor refinement. However, when applied to incomplete views, they still face two major challenges: unstable anchor selection and limited utilization of high-order information. These limitations degrade the quality of embedding features and affect clustering performance. To address these challenges, we propose IMVC-TPAL (Efficient Incomplete Multi-View Tensor Clustering through Predefined Anchor Learning) , which begins with a customized anchor selection strategy that reduces randomness and mitigates the impact of missing views, and further incorporates adaptive anchor graph completion directly into the embedding learning process. Additionally, a tensor-based low-frequency approximation operator is employed to explore intra-view similarity, resulting in smooth and discriminative embedding features. In experiments conducted on five datasets under three missing-view ratios, IMVC-TPAL achieves the best performance on 73.3% of all evaluation metrics and ranks second on the remaining ones, demonstrating its effectiveness. These results confirm that our method successfully integrates predefined-anchor learning with the incomplete multi-view setting, providing a reliable and scalable solution for IMVC.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"191 ","pages":"Article 114660"},"PeriodicalIF":6.6,"publicationDate":"2026-01-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146039852","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Combinatorial optimization problems (COPs) present significant computational challenges due to their discrete nature, increasing complexity, and NP-hard characteristics. Identifying an effective solver is particularly difficult given the large variety of existing techniques, including exact algorithms, metaheuristics, and neural-network-based approaches such as Hopfield networks (HNs). Although HNs have shown strong potential for solving complex COPs through an energy-minimization framework, their performance is highly sensitive to the choice of hyperparameters and the initialization strategy, both of which require careful tuning. This paper introduces a new method that enhances the effectiveness of HNs for COPs by jointly optimizing their hyperparameters and starting point using the Arithmetic Optimization Algorithm (AOA). The goal is to develop a recurrent-neural-network-based approach that leverages systematic hyperparameter tuning and optimal initialization to improve solution quality and convergence behavior. Experimental results demonstrate that the proposed method achieves optimal solutions on 20 instances of the task assignment problem (TAP) and provides high-quality solutions for the graph coloring problem (GCP) and the traveling salesman problem (TSP) within reasonable computational times. Compared to a genetic algorithm (GA) and traditional HNs with random hyperparameter selection, the proposed approach achieves performance improvements of 52.83% for TAP, 28.97% for GCP, and 9.32% for TSP.
{"title":"Hopfield network-based algorithm for combinatorial optimization","authors":"Houssam Hamdouch , Safae Rbihou , Kaoutar Senhaji , Khalid Haddouch","doi":"10.1016/j.asoc.2026.114652","DOIUrl":"10.1016/j.asoc.2026.114652","url":null,"abstract":"<div><div>Combinatorial optimization problems (COPs) present significant computational challenges due to their discrete nature, increasing complexity, and NP-hard characteristics. Identifying an effective solver is particularly difficult given the large variety of existing techniques, including exact algorithms, metaheuristics, and neural-network-based approaches such as Hopfield networks (HNs). Although HNs have shown strong potential for solving complex COPs through an energy-minimization framework, their performance is highly sensitive to the choice of hyperparameters and the initialization strategy, both of which require careful tuning. This paper introduces a new method that enhances the effectiveness of HNs for COPs by jointly optimizing their hyperparameters and starting point using the Arithmetic Optimization Algorithm (AOA). The goal is to develop a recurrent-neural-network-based approach that leverages systematic hyperparameter tuning and optimal initialization to improve solution quality and convergence behavior. Experimental results demonstrate that the proposed method achieves optimal solutions on 20 instances of the task assignment problem (TAP) and provides high-quality solutions for the graph coloring problem (GCP) and the traveling salesman problem (TSP) within reasonable computational times. Compared to a genetic algorithm (GA) and traditional HNs with random hyperparameter selection, the proposed approach achieves performance improvements of 52.83% for TAP, 28.97% for GCP, and 9.32% for TSP.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"190 ","pages":"Article 114652"},"PeriodicalIF":6.6,"publicationDate":"2026-01-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145979789","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-01-13DOI: 10.1016/j.asoc.2026.114599
Ibrahim Yousef Alshareef , Ab Al-Hadi Ab Rahman , Nuzhat Khan , Hasan Alqaraghuli
Spectral convolutional neural networks using Fast Fourier Transform (FFT) often suffer from high computational complexity and memory demands due to complex valued operations and the need for inverse transforms limiting their deployment on resource constrained devices. This paper presents a novel end-to-end spectral convolutional neural network (SpCNN) architecture that operates entirely in the Discrete Cosine Transform (DCT) domain, eliminating the need for inverse transformations and complex arithmetic. Leveraging the DCT’s real valued representation and superior energy compaction, the proposed design significantly reduces computational workload and memory usage while preserving classification accuracy. Key innovations include the removal of IFFT layers, a frequency domain adaptive activation function (FReLU), and a DCT optimized spectral pooling mechanism, each tailored for deployment in low power, resource constrained environments. Experimental evaluations on MNIST and a 94-class ASCII dataset demonstrate the model’s efficiency: LeNet5-DCT achieves a 37.96% FLOPs reduction, 18.45% lower memory usage, and 96.56% test accuracy, while VGG7-DCT achieves a 33.95% FLOPs reduction, 14.32% lower memory usage, and 90.62% test accuracy. The architecture also shows strong robustness to quantization, confirming its suitability for edge AI applications and low energy inference. This work provides a scalable, hardware efficient spectral learning framework, paving the way for future hybrid spectral models optimized for embedded environments.
{"title":"End-to-end discrete cosine transform integration in spectral convolutional neural networks for resource-efficient deep learning","authors":"Ibrahim Yousef Alshareef , Ab Al-Hadi Ab Rahman , Nuzhat Khan , Hasan Alqaraghuli","doi":"10.1016/j.asoc.2026.114599","DOIUrl":"10.1016/j.asoc.2026.114599","url":null,"abstract":"<div><div>Spectral convolutional neural networks using Fast Fourier Transform (FFT) often suffer from high computational complexity and memory demands due to complex valued operations and the need for inverse transforms limiting their deployment on resource constrained devices. This paper presents a novel end-to-end spectral convolutional neural network (SpCNN) architecture that operates entirely in the Discrete Cosine Transform (DCT) domain, eliminating the need for inverse transformations and complex arithmetic. Leveraging the DCT’s real valued representation and superior energy compaction, the proposed design significantly reduces computational workload and memory usage while preserving classification accuracy. Key innovations include the removal of IFFT layers, a frequency domain adaptive activation function (FReLU), and a DCT optimized spectral pooling mechanism, each tailored for deployment in low power, resource constrained environments. Experimental evaluations on MNIST and a 94-class ASCII dataset demonstrate the model’s efficiency: LeNet5-DCT achieves a 37.96% FLOPs reduction, 18.45% lower memory usage, and 96.56% test accuracy, while VGG7-DCT achieves a 33.95% FLOPs reduction, 14.32% lower memory usage, and 90.62% test accuracy. The architecture also shows strong robustness to quantization, confirming its suitability for edge AI applications and low energy inference. This work provides a scalable, hardware efficient spectral learning framework, paving the way for future hybrid spectral models optimized for embedded environments.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"191 ","pages":"Article 114599"},"PeriodicalIF":6.6,"publicationDate":"2026-01-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146039850","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-01-13DOI: 10.1016/j.asoc.2026.114650
Zixuan Zhang , Fan Shi , Chen Jia , Mianzhao Wang , Xu Cheng
Extracting shared boundary cues across different anomaly domains is critical to enhancing generalization on unseen data, thereby laying the foundation for unified industrial anomaly detection paradigms. Existing unified detection paradigms often directly extract discriminative features from multiple data domains. However, due to the inherent semantic gaps between different data sources, bridging this disparity within a shared feature representation across multiple data domains remains a key challenge. To address this challenge, we propose a boundary-guided large-scale vision model that extracts commonalities across diverse domains. Specifically, we generate initial feature embeddings by establishing a multi-domain normal sample repository and employing a parameter coupling strategy. This captures shared boundary information across different data domains, thereby reducing the inherent semantic gaps. For anomalous feature synthesis, we incorporate this boundary information into the generation process, ensuring that the synthesized features retain critical structural details while expanding the coverage of potential anomalous data distributions. Additionally, to enhance feature space separation between normal and anomalous samples, we introduce a hybrid constraint optimization mechanism that improves the discriminative ability of the model. Extensive experiments on the MVTec AD, VisA, and MPDD datasets demonstrate that our method achieves state-of-the-art performance across various industrial scenarios. Experimental results demonstrate the effectiveness of boundary-guided shared information for multi-domain anomaly detection.
{"title":"Boundary-guided large-scale vision model for unified multi-domain industrial anomaly detection","authors":"Zixuan Zhang , Fan Shi , Chen Jia , Mianzhao Wang , Xu Cheng","doi":"10.1016/j.asoc.2026.114650","DOIUrl":"10.1016/j.asoc.2026.114650","url":null,"abstract":"<div><div>Extracting shared boundary cues across different anomaly domains is critical to enhancing generalization on unseen data, thereby laying the foundation for unified industrial anomaly detection paradigms. Existing unified detection paradigms often directly extract discriminative features from multiple data domains. However, due to the inherent semantic gaps between different data sources, bridging this disparity within a shared feature representation across multiple data domains remains a key challenge. To address this challenge, we propose a boundary-guided large-scale vision model that extracts commonalities across diverse domains. Specifically, we generate initial feature embeddings by establishing a multi-domain normal sample repository and employing a parameter coupling strategy. This captures shared boundary information across different data domains, thereby reducing the inherent semantic gaps. For anomalous feature synthesis, we incorporate this boundary information into the generation process, ensuring that the synthesized features retain critical structural details while expanding the coverage of potential anomalous data distributions. Additionally, to enhance feature space separation between normal and anomalous samples, we introduce a hybrid constraint optimization mechanism that improves the discriminative ability of the model. Extensive experiments on the MVTec AD, VisA, and MPDD datasets demonstrate that our method achieves state-of-the-art performance across various industrial scenarios. Experimental results demonstrate the effectiveness of boundary-guided shared information for multi-domain anomaly detection.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"191 ","pages":"Article 114650"},"PeriodicalIF":6.6,"publicationDate":"2026-01-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146039851","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
扫码关注我们
求助内容：
应助结果提醒方式：
京公网安备 11010802042870号