Pub Date : 2025-03-05DOI: 10.1016/j.ins.2025.122050
Wanzheng Qiu , JinRong Wang , Dong Shen
In this paper, the consensus problem of multi-agent systems with unknown faded neighborhood information is addressed using the iterative learning control method. Considering that information exchange in wireless networks may be disturbed by unknown fading effects and unknown additive noise, it is significant to realize accurate consensus tracking of each agent to a given leader under the contaminated information. Unlike the traditional mechanism of correcting unknown faded neighborhood information by estimating the statistical characteristics of random fading variables, we introduce test signals to correct the trajectory signals of each agent. As no estimation mechanism is involved, the storage and computational burden of the whole system are greatly reduced. Based on a classic distributed structure and a novel correction mechanism, two novel distributed learning consensus control schemes are constructed. The consensus results of multi-agent systems under the two learning control schemes are discussed in detail using mathematical analysis tools. Finally, the multi-pendulum network system is simulated to verify the theoretical results.
{"title":"Consensus control for multi-agent systems with unknown faded neighborhood information via iterative learning scheme","authors":"Wanzheng Qiu , JinRong Wang , Dong Shen","doi":"10.1016/j.ins.2025.122050","DOIUrl":"10.1016/j.ins.2025.122050","url":null,"abstract":"<div><div>In this paper, the consensus problem of multi-agent systems with unknown faded neighborhood information is addressed using the iterative learning control method. Considering that information exchange in wireless networks may be disturbed by unknown fading effects and unknown additive noise, it is significant to realize accurate consensus tracking of each agent to a given leader under the contaminated information. Unlike the traditional mechanism of correcting unknown faded neighborhood information by estimating the statistical characteristics of random fading variables, we introduce test signals to correct the trajectory signals of each agent. As no estimation mechanism is involved, the storage and computational burden of the whole system are greatly reduced. Based on a classic distributed structure and a novel correction mechanism, two novel distributed learning consensus control schemes are constructed. The consensus results of multi-agent systems under the two learning control schemes are discussed in detail using mathematical analysis tools. Finally, the multi-pendulum network system is simulated to verify the theoretical results.</div></div>","PeriodicalId":51063,"journal":{"name":"Information Sciences","volume":"708 ","pages":"Article 122050"},"PeriodicalIF":8.1,"publicationDate":"2025-03-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143561984","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 : 2025-03-05DOI: 10.1016/j.ins.2025.122062
Sumin Yu , Jia Xiao , Zhijiao Du , Xuanhua Xu
Social network large-scale decision-making (SNLSDM) has become an important domain in the field of decision science. A major challenge in solving such problems lies in the effective data dimensionality reduction through clustering techniques. While the reliability of evaluation information significantly influences clustering quality, most clustering algorithms overlook this critical factor. To address this gap, this article proposes novel reliability-driven joint clustering algorithms based on hybrid attribute analysis for SNLSDM problems. First, the probabilistic linguistic evaluation-reliability function (PL-ERF) is defined to handle fuzzy evaluations and social networks, along with its operational rules. We describe the configuration of SNLSDM with PL-ERFs. Subsequently, the hybrid attribute analysis is conducted to process the initial trust social network. A new reliability-based trust propagation method is designed to construct a complete trust social network. By combining trust and similarity information, a compatibility network is established. Accordingly, we develop reliability-driven joint clustering algorithms that consider multiple constraints, including similarity, trust, and compatibility. We also discuss the time complexity analysis and scalability of the algorithms. Finally, a numerical experiment and two real-word cases studies illustrate the feasibility and effectiveness of the algorithms. A comparative analysis highlights the impact and advantages of incorporating reliability into clustering.
{"title":"Reliability-driven joint clustering based on hybrid attribute analysis for supporting social network large-scale decision-making","authors":"Sumin Yu , Jia Xiao , Zhijiao Du , Xuanhua Xu","doi":"10.1016/j.ins.2025.122062","DOIUrl":"10.1016/j.ins.2025.122062","url":null,"abstract":"<div><div>Social network large-scale decision-making (SNLSDM) has become an important domain in the field of decision science. A major challenge in solving such problems lies in the effective data dimensionality reduction through clustering techniques. While the reliability of evaluation information significantly influences clustering quality, most clustering algorithms overlook this critical factor. To address this gap, this article proposes novel reliability-driven joint clustering algorithms based on hybrid attribute analysis for SNLSDM problems. First, the probabilistic linguistic evaluation-reliability function (PL-ERF) is defined to handle fuzzy evaluations and social networks, along with its operational rules. We describe the configuration of SNLSDM with PL-ERFs. Subsequently, the hybrid attribute analysis is conducted to process the initial trust social network. A new reliability-based trust propagation method is designed to construct a complete trust social network. By combining trust and similarity information, a compatibility network is established. Accordingly, we develop reliability-driven joint clustering algorithms that consider multiple constraints, including similarity, trust, and compatibility. We also discuss the time complexity analysis and scalability of the algorithms. Finally, a numerical experiment and two real-word cases studies illustrate the feasibility and effectiveness of the algorithms. A comparative analysis highlights the impact and advantages of incorporating reliability into clustering.</div></div>","PeriodicalId":51063,"journal":{"name":"Information Sciences","volume":"709 ","pages":"Article 122062"},"PeriodicalIF":8.1,"publicationDate":"2025-03-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143593712","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 : 2025-03-05DOI: 10.1016/j.ins.2025.122045
Chunmao Jiang , Guojun Mao , Bin Xie
Container-based cloud computing requires efficient and adaptive resource management, particularly when making vertical scaling decisions. Traditional approaches often struggle with workload variability and lack flexibility when faced with uncertainties in workload patterns. This paper introduces a novel three-way decision-based reinforcement learning (TWD-RL) model for container vertical scaling. The TWD-RL model partitions the state space into positive, boundary, and negative regions based on confidence measures derived from historical data and current system states. This partitioning enables more nuanced scaling decisions: immediate scaling in high-confidence states, deferring decisions in uncertain states, and exploring in low-confidence states. We provide a theoretical analysis of the model's convergence properties and optimality conditions, thus establishing its mathematical foundation. Furthermore, we evaluate our model using real-world workload data from the Google Cloud Platform. The results demonstrate that TWD-RL significantly outperforms traditional Vertical Pod Autoscaler (VPA) approaches with respect to average response time, Service Level Agreement (SLA) violations, and resource utilization efficiency.
{"title":"Three-way decision-based reinforcement learning for container vertical scaling","authors":"Chunmao Jiang , Guojun Mao , Bin Xie","doi":"10.1016/j.ins.2025.122045","DOIUrl":"10.1016/j.ins.2025.122045","url":null,"abstract":"<div><div>Container-based cloud computing requires efficient and adaptive resource management, particularly when making vertical scaling decisions. Traditional approaches often struggle with workload variability and lack flexibility when faced with uncertainties in workload patterns. This paper introduces a novel three-way decision-based reinforcement learning (TWD-RL) model for container vertical scaling. The TWD-RL model partitions the state space into positive, boundary, and negative regions based on confidence measures derived from historical data and current system states. This partitioning enables more nuanced scaling decisions: immediate scaling in high-confidence states, deferring decisions in uncertain states, and exploring in low-confidence states. We provide a theoretical analysis of the model's convergence properties and optimality conditions, thus establishing its mathematical foundation. Furthermore, we evaluate our model using real-world workload data from the Google Cloud Platform. The results demonstrate that TWD-RL significantly outperforms traditional Vertical Pod Autoscaler (VPA) approaches with respect to average response time, Service Level Agreement (SLA) violations, and resource utilization efficiency.</div></div>","PeriodicalId":51063,"journal":{"name":"Information Sciences","volume":"708 ","pages":"Article 122045"},"PeriodicalIF":8.1,"publicationDate":"2025-03-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143549450","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 : 2025-03-04DOI: 10.1016/j.ins.2025.122042
Yiran Yu , Dewei Li , Baoming Han , Qi Zhang , Yue Huang , Ruixia Yang
This paper proposed a travel plan recommendation system that can provide multi-modal, personalized, and door-to-door travel plans to solve travelers’ difficulty in choosing when facing vast and complex travel information. First, we established a dynamic Travel Choice Behavior Graph (TCBG) model, which considers the travel plan candidate set and the temporal characteristics (time-decay and periodicity) of travelers’ behavioral preferences. Next, to effectively learn from TCBG, we constructed a Unified Candidate Set Representation Module (UCSRM) and a new graph neural network called Continuous Dynamic Heterogeneous Graph Attention Networks (CDHAN). UCSRM can employ a multi-head self-attention mechanism for a unified representation of travel plan candidate sets with inconsistent lengths. CDHAN can capture the temporal characteristics of travelers’ preferences by combining the improved Hawkes process. Finally, we validated the effectiveness of the model and framework on multi-modal travel datasets and achieved 0.8172, 0.7994, 0.7859, and 0.9345 on the evaluation metrics of Pre, Rec, F1, and NDCG, respectively. These results show that our model/framework outperforms six existing state-of-the-art models/frameworks in these four evaluation metrics. This study provided a new model and learning framework for travel plan recommendation systems, essential for improving the efficiency of urban transportation and travelers’ travel experience.
{"title":"Enhancing urban mobility: A multi-modal travel plan recommendation framework integrating the influences of temporal characteristics and candidate sets","authors":"Yiran Yu , Dewei Li , Baoming Han , Qi Zhang , Yue Huang , Ruixia Yang","doi":"10.1016/j.ins.2025.122042","DOIUrl":"10.1016/j.ins.2025.122042","url":null,"abstract":"<div><div>This paper proposed a travel plan recommendation system that can provide multi-modal, personalized, and door-to-door travel plans to solve travelers’ difficulty in choosing when facing vast and complex travel information. First, we established a dynamic <strong>T</strong>ravel <strong>C</strong>hoice <strong>B</strong>ehavior <strong>G</strong>raph (TCBG) model, which considers the travel plan candidate set and the temporal characteristics (time-decay and periodicity) of travelers’ behavioral preferences. Next, to effectively learn from TCBG, we constructed a <strong>U</strong>nified <strong>C</strong>andidate <strong>S</strong>et <strong>R</strong>epresentation <strong>M</strong>odule (UCSRM) and a new graph neural network called <strong>C</strong>ontinuous <strong>D</strong>ynamic <strong>H</strong>eterogeneous Graph <strong>A</strong>ttention <strong>N</strong>etworks (CDHAN). UCSRM can employ a multi-head self-attention mechanism for a unified representation of travel plan candidate sets with inconsistent lengths. CDHAN can capture the temporal characteristics of travelers’ preferences by combining the improved Hawkes process. Finally, we validated the effectiveness of the model and framework on multi-modal travel datasets and achieved 0.8172, 0.7994, 0.7859, and 0.9345 on the evaluation metrics of Pre, Rec, F1, and NDCG, respectively. These results show that our model/framework outperforms six existing state-of-the-art models/frameworks in these four evaluation metrics. This study provided a new model and learning framework for travel plan recommendation systems, essential for improving the efficiency of urban transportation and travelers’ travel experience.</div></div>","PeriodicalId":51063,"journal":{"name":"Information Sciences","volume":"709 ","pages":"Article 122042"},"PeriodicalIF":8.1,"publicationDate":"2025-03-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143593661","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 : 2025-03-04DOI: 10.1016/j.ins.2025.122052
Yuanshan Liu, Yude Xia
This paper employs data-driven techniques to investigate the robustness control of leader-follower consensus in nonlinear discrete-time time-varying multiagent systems with fixed topology. Initially, pertinent symbolic definitions for sampled data are established, followed by an introduction to graph theory and system models. As data-driven algorithms necessitate linear systems, each nonlinear subsystem is linearized. Subsequently, distributed controllers are designed based on control principles to ensure multi-agent consensus. Additionally, the controller gain matrix is derived via a data-driven method, with its feasibility theoretically verified by solving nonlinear matrix inequalities. Finally, numerical simulations validate the efficacy of this approach for achieving robust leader-follower consensus control.
{"title":"Investigation of consensus for nonlinear time-varying multiagent systems via data-driven techniques","authors":"Yuanshan Liu, Yude Xia","doi":"10.1016/j.ins.2025.122052","DOIUrl":"10.1016/j.ins.2025.122052","url":null,"abstract":"<div><div>This paper employs data-driven techniques to investigate the robustness control of leader-follower consensus in nonlinear discrete-time time-varying multiagent systems with fixed topology. Initially, pertinent symbolic definitions for sampled data are established, followed by an introduction to graph theory and system models. As data-driven algorithms necessitate linear systems, each nonlinear subsystem is linearized. Subsequently, distributed controllers are designed based on control principles to ensure multi-agent consensus. Additionally, the controller gain matrix is derived via a data-driven method, with its feasibility theoretically verified by solving nonlinear matrix inequalities. Finally, numerical simulations validate the efficacy of this approach for achieving robust leader-follower consensus control.</div></div>","PeriodicalId":51063,"journal":{"name":"Information Sciences","volume":"708 ","pages":"Article 122052"},"PeriodicalIF":8.1,"publicationDate":"2025-03-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143549449","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 : 2025-03-04DOI: 10.1016/j.ins.2025.122044
Mostafa K. El-Bably , Rodyna A. Hosny , Mostafa A. El-Gayar
The theory of rough sets produces a potent framework for administrating uncertainty and ambiguity in data, which is crucial for effective decision-making. However, the reliance on equivalence relations within this framework has led to the exploration of various generalizations and extensions. In this paper, we introduce eight new types of initial neighborhoods, expanding on the idea of initial neighborhoods, and examine the relationships and properties of twelve distinct types of neighborhoods derived from binary relations. We define initial-minimal and initial-maximal neighborhoods and develop eight types of rough approximations (-approximations) that generalize Pawlak's theory. These new approximations significantly improve upon previous methods, achieving accuracy rates of up to 100%. Furthermore, we implement Generalized Nano-topological frameworks in conjunction with our novel methodologies to address clinical applications, particularly focusing on advancing diagnostic strategies for Covid-19. By employing a universal binary relation, we clarify the effectiveness for our methodology per enhancing decision-making processes and pinpointing significant risk factors associated with Covid-19. Additionally, we introduce two algorithms for decision-making problems in information systems, emphasizing the broader applicability and significance of our approach across various fields.
{"title":"Innovative rough set approaches using novel initial-neighborhood systems: Applications in medical diagnosis of Covid-19 variants","authors":"Mostafa K. El-Bably , Rodyna A. Hosny , Mostafa A. El-Gayar","doi":"10.1016/j.ins.2025.122044","DOIUrl":"10.1016/j.ins.2025.122044","url":null,"abstract":"<div><div>The theory of rough sets produces a potent framework for administrating uncertainty and ambiguity in data, which is crucial for effective decision-making. However, the reliance on equivalence relations within this framework has led to the exploration of various generalizations and extensions. In this paper, we introduce eight new types of initial neighborhoods, expanding on the idea of initial neighborhoods, and examine the relationships and properties of twelve distinct types of neighborhoods derived from binary relations. We define initial-minimal and initial-maximal neighborhoods and develop eight types of rough approximations (<span><math><msub><mrow><mi>I</mi></mrow><mrow><mi>ȷ</mi></mrow></msub></math></span>-approximations) that generalize Pawlak's theory. These new approximations significantly improve upon previous methods, achieving accuracy rates of up to 100%. Furthermore, we implement Generalized Nano-topological frameworks in conjunction with our novel methodologies to address clinical applications, particularly focusing on advancing diagnostic strategies for Covid-19. By employing a universal binary relation, we clarify the effectiveness for our methodology per enhancing decision-making processes and pinpointing significant risk factors associated with Covid-19. Additionally, we introduce two algorithms for decision-making problems in information systems, emphasizing the broader applicability and significance of our approach across various fields.</div></div>","PeriodicalId":51063,"journal":{"name":"Information Sciences","volume":"708 ","pages":"Article 122044"},"PeriodicalIF":8.1,"publicationDate":"2025-03-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143561981","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 : 2025-03-03DOI: 10.1016/j.ins.2025.122043
Han Wang , Yanbing Ju , Yongxing Chang , Enrique Herrera-Viedma
The futures portfolio is a key tool for addressing market volatility and complexity in the financial markets. Traditional static strategies struggle to keep up with the rapidly shifting market sentiment and herd behavior, leading to delayed decision-making and risk management failures. To enhance investment efficiency and improve risk control, we propose a dynamic multi-criteria nested sequential three-state three-way decision (TS3WD) model based on herd behavior to identify and implement herd behaviors and optimize the futures portfolio strategy. Firstly, this paper proposes a method for determining optimistic and pessimistic conditional probabilities based on loss functions, deriving new TS3WD and simplified decision rules. Secondly, the herd behavior discrimination method is introduced to divide it into positive, neutral, and negative herd behaviors for holding futures contracts. Thirdly, four minimum adjustment optimization models for positive and negative herd behaviors under optimistic and pessimistic attitudes are constructed based on new decision rules, respectively, and a method based on the self-confidence principle for neutral herd behavior is presented, providing a quantitative model for implementing herd behaviors. Subsequently, a progressive dynamic algorithm based on a multi-criteria nested sequential TS3WD model is proposed to deduce the futures portfolio strategy, which dynamically identifies and adjusts loss functions to obtain the optimal futures investment behavior, forming a complete futures portfolio strategy. Finally, we apply the proposed method to solve the metal futures portfolio strategy in the Shanghai Futures Exchange, providing implications for investors in the futures market through sensitivity and comparative analyses.
{"title":"Dynamic futures portfolio strategy: A multi-criteria nested sequential three-state three-way decision model based on herd behavior","authors":"Han Wang , Yanbing Ju , Yongxing Chang , Enrique Herrera-Viedma","doi":"10.1016/j.ins.2025.122043","DOIUrl":"10.1016/j.ins.2025.122043","url":null,"abstract":"<div><div>The futures portfolio is a key tool for addressing market volatility and complexity in the financial markets. Traditional static strategies struggle to keep up with the rapidly shifting market sentiment and herd behavior, leading to delayed decision-making and risk management failures. To enhance investment efficiency and improve risk control, we propose a dynamic multi-criteria nested sequential three-state three-way decision (TS3WD) model based on herd behavior to identify and implement herd behaviors and optimize the futures portfolio strategy. Firstly, this paper proposes a method for determining optimistic and pessimistic conditional probabilities based on loss functions, deriving new TS3WD and simplified decision rules. Secondly, the herd behavior discrimination method is introduced to divide it into positive, neutral, and negative herd behaviors for holding futures contracts. Thirdly, four minimum adjustment optimization models for positive and negative herd behaviors under optimistic and pessimistic attitudes are constructed based on new decision rules, respectively, and a method based on the self-confidence principle for neutral herd behavior is presented, providing a quantitative model for implementing herd behaviors. Subsequently, a progressive dynamic algorithm based on a multi-criteria nested sequential TS3WD model is proposed to deduce the futures portfolio strategy, which dynamically identifies and adjusts loss functions to obtain the optimal futures investment behavior, forming a complete futures portfolio strategy. Finally, we apply the proposed method to solve the metal futures portfolio strategy in the Shanghai Futures Exchange, providing implications for investors in the futures market through sensitivity and comparative analyses.</div></div>","PeriodicalId":51063,"journal":{"name":"Information Sciences","volume":"708 ","pages":"Article 122043"},"PeriodicalIF":8.1,"publicationDate":"2025-03-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143549448","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 : 2025-03-03DOI: 10.1016/j.ins.2025.122046
Lan Huang , Yangguang Shao , Wenju Hou , Hui Yang , Yan Wang , Nan Sheng , Yinglu Sun , Yao Wang
Brain tumors are among the most prevalent and deadly diseases worldwide, making early diagnosis critical. However, existing automated brain tumor diagnostic methods often lack interpretability, and the high cost of labeled data limits their effectiveness. Class activation mapping (CAM) provides visual explanations and object localization for convolutional neural networks (CNNs) by highlighting regions of interest corresponding to specific classes. However, existing approaches tend to focus solely on discriminative regions and often contain excessive noise. In this paper, we propose a simpler and more efficient method called PCG-CAM, which provides visual explanations for brain tumor diagnosis and generates fine-grained pseudo-labels. PCG-CAM extracts the principal components of gradients and uses their absolute values as weights for the feature maps, thereby better reflecting the importance of each feature map while preserving more object features. We evaluated the saliency maps generated by PCG-CAM on weakly-supervised brain tumor segmentation and assessed their generalizability in object localization tasks. Specifically, our method achieves 47.42% mIoU in weakly-supervised brain tumor segmentation, outperforming other methods by nearly 10% on average. The results on brain MRI and natural images demonstrate that our method effectively localizes target positions and provides robust explanations for model decisions.
{"title":"PCG-CAM: Enhanced class activation map using principal components of gradients and its applications in brain MRI","authors":"Lan Huang , Yangguang Shao , Wenju Hou , Hui Yang , Yan Wang , Nan Sheng , Yinglu Sun , Yao Wang","doi":"10.1016/j.ins.2025.122046","DOIUrl":"10.1016/j.ins.2025.122046","url":null,"abstract":"<div><div>Brain tumors are among the most prevalent and deadly diseases worldwide, making early diagnosis critical. However, existing automated brain tumor diagnostic methods often lack interpretability, and the high cost of labeled data limits their effectiveness. Class activation mapping (CAM) provides visual explanations and object localization for convolutional neural networks (CNNs) by highlighting regions of interest corresponding to specific classes. However, existing approaches tend to focus solely on discriminative regions and often contain excessive noise. In this paper, we propose a simpler and more efficient method called PCG-CAM, which provides visual explanations for brain tumor diagnosis and generates fine-grained pseudo-labels. PCG-CAM extracts the principal components of gradients and uses their absolute values as weights for the feature maps, thereby better reflecting the importance of each feature map while preserving more object features. We evaluated the saliency maps generated by PCG-CAM on weakly-supervised brain tumor segmentation and assessed their generalizability in object localization tasks. Specifically, our method achieves 47.42% mIoU in weakly-supervised brain tumor segmentation, outperforming other methods by nearly 10% on average. The results on brain MRI and natural images demonstrate that our method effectively localizes target positions and provides robust explanations for model decisions.</div></div>","PeriodicalId":51063,"journal":{"name":"Information Sciences","volume":"708 ","pages":"Article 122046"},"PeriodicalIF":8.1,"publicationDate":"2025-03-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143579088","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 : 2025-03-01DOI: 10.1016/j.ins.2025.122036
Yu Gao , Huaiping Jin , Zhiqiang Wang , Bin Wang , Bin Qian , Biao Yang
Deep learning techniques have been widely applied for industrial quality prediction. However, industrial process data are often generated as data streams, which typically exhibit characteristics such as strong nonlinearity, time-varying behavior, and low sampling rates of quality variables. Conventional offline-trained deep learning models often fail to provide accurate predictions for such semi-supervised data streams. Therefore, a quality-relevant deep rule-based system with complementary lifelong learning (QDRSCLL) is proposed to enable adaptive prediction of critical quality variables in streaming data environments. QDRSCLL comprises a deep backbone network and a shallow predictor. The former utilizes a semi-supervised quality-relevant stacked autoencoder (SQSAE) for feature extraction, while the latter employs a hierarchical fuzzy rule system (HFRS) to perform fuzzy inference on hierarchical hidden features. Furthermore, a novel complementary lifelong learning mechanism is proposed to enable QDRSCLL with online incremental learning capabilities. Additionally, semi-supervised learning is integrated into the online learning process to further enhance its deep feature extraction capabilities and the prediction performance. The feasibility and superiority of the proposed method are demonstrated through two real-world processes and four synthetic datasets. Compared to the traditional evolving fuzzy system (EFS), the RMSE of QDRSCLL is reduced by more than 25% in all application scenarios.
{"title":"A quality-relevant deep rule-based system with complementary lifelong learning for adaptive quality prediction in industrial semi-supervised process data streams","authors":"Yu Gao , Huaiping Jin , Zhiqiang Wang , Bin Wang , Bin Qian , Biao Yang","doi":"10.1016/j.ins.2025.122036","DOIUrl":"10.1016/j.ins.2025.122036","url":null,"abstract":"<div><div>Deep learning techniques have been widely applied for industrial quality prediction. However, industrial process data are often generated as data streams, which typically exhibit characteristics such as strong nonlinearity, time-varying behavior, and low sampling rates of quality variables. Conventional offline-trained deep learning models often fail to provide accurate predictions for such semi-supervised data streams. Therefore, a quality-relevant deep rule-based system with complementary lifelong learning (QDRSCLL) is proposed to enable adaptive prediction of critical quality variables in streaming data environments. QDRSCLL comprises a deep backbone network and a shallow predictor. The former utilizes a semi-supervised quality-relevant stacked autoencoder (SQSAE) for feature extraction, while the latter employs a hierarchical fuzzy rule system (HFRS) to perform fuzzy inference on hierarchical hidden features. Furthermore, a novel complementary lifelong learning mechanism is proposed to enable QDRSCLL with online incremental learning capabilities. Additionally, semi-supervised learning is integrated into the online learning process to further enhance its deep feature extraction capabilities and the prediction performance. The feasibility and superiority of the proposed method are demonstrated through two real-world processes and four synthetic datasets. Compared to the traditional evolving fuzzy system (EFS), the RMSE of QDRSCLL is reduced by more than 25% in all application scenarios.</div></div>","PeriodicalId":51063,"journal":{"name":"Information Sciences","volume":"708 ","pages":"Article 122036"},"PeriodicalIF":8.1,"publicationDate":"2025-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143579089","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 : 2025-02-28DOI: 10.1016/j.ins.2025.122039
Rong Fei , Yuxin Wan , Bo Hu , Aimin Li , Yingan Cui , Hailong Peng
The incomplete network is defined as the network with missing edges, which forms incomplete network topology by missing real information because of multiple-factor such as personal privacy security and threats, etc. Academic interest in incomplete network studies is increasing. Some methods solving community detection problem in the incomplete network, as link prediction, show low ACC or NMI. To address those, there is a need for approaches less affected by missing edges and easy to obtain communities. We propose a deep core node information embedding(DCNIE) algorithm on network with missing edges for community detection, aiming to obtain core node information rather than the influence of edges. First, by edge augmentation, the network with missing edges is integrated into complete networks. Second, the k-core algorithm is used to obtain core node information and build a similarity matrix, followed by an unsupervised deep method that implements network embedding to obtain a low-dimensional feature matrix. Finally, Gaussian mixture model is used for clustering to obtain the community division. We compare eleven state-of-the-art methods on eleven real networks by using eight evaluation metrics. Experiments demonstrate that DCNIE is superior in performance and efficiency while gaining accurate community division in incomplete network.
{"title":"Deep core node information embedding on networks with missing edges for community detection","authors":"Rong Fei , Yuxin Wan , Bo Hu , Aimin Li , Yingan Cui , Hailong Peng","doi":"10.1016/j.ins.2025.122039","DOIUrl":"10.1016/j.ins.2025.122039","url":null,"abstract":"<div><div>The incomplete network is defined as the network with missing edges, which forms incomplete network topology by missing real information because of multiple-factor such as personal privacy security and threats, etc. Academic interest in incomplete network studies is increasing. Some methods solving community detection problem in the incomplete network, as link prediction, show low ACC or NMI. To address those, there is a need for approaches less affected by missing edges and easy to obtain communities. We propose a deep core node information embedding(DCNIE) algorithm on network with missing edges for community detection, aiming to obtain core node information rather than the influence of edges. First, by edge augmentation, the network with missing edges is integrated into complete networks. Second, the <em>k</em>-core algorithm is used to obtain core node information and build a similarity matrix, followed by an unsupervised deep method that implements network embedding to obtain a low-dimensional feature matrix. Finally, Gaussian mixture model is used for clustering to obtain the community division. We compare eleven state-of-the-art methods on eleven real networks by using eight evaluation metrics. Experiments demonstrate that DCNIE is superior in performance and efficiency while gaining accurate community division in incomplete network.</div></div>","PeriodicalId":51063,"journal":{"name":"Information Sciences","volume":"707 ","pages":"Article 122039"},"PeriodicalIF":8.1,"publicationDate":"2025-02-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143529497","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