Pub Date : 2025-03-07DOI: 10.1016/j.ins.2025.122070
Ying Wang , Yuanhua Wang , Yonglin Guo , Wenke Zang , Guodong Zhao
The formation of coalitions or alliances is ubiquitous in nature. This paper extends the zero-determinant (ZD) strategies from single players to subsets of players, to which we refer to as the ZD alliances. First, we model the dynamics of repeated multi-player games with a strategic alliance as an equivalent algebraic form and a simple formula is proposed to design a strategic ZD alliance. Thereafter, the modeling and designing methods for synchronous ZD alliance are considered. Finally, we apply the main results to a four-player prisoner's dilemma and analyze the influence of parameters on the probability of alliance cooperation. Our results can not only reduce the computation complexity, but also highlight the importance of coordination to succeed in large groups.
{"title":"Design of zero-determinant alliances in repeated multi-player games","authors":"Ying Wang , Yuanhua Wang , Yonglin Guo , Wenke Zang , Guodong Zhao","doi":"10.1016/j.ins.2025.122070","DOIUrl":"10.1016/j.ins.2025.122070","url":null,"abstract":"<div><div>The formation of coalitions or alliances is ubiquitous in nature. This paper extends the zero-determinant (ZD) strategies from single players to subsets of players, to which we refer to as the ZD alliances. First, we model the dynamics of repeated multi-player games with a strategic alliance as an equivalent algebraic form and a simple formula is proposed to design a strategic ZD alliance. Thereafter, the modeling and designing methods for synchronous ZD alliance are considered. Finally, we apply the main results to a four-player prisoner's dilemma and analyze the influence of parameters on the probability of alliance cooperation. Our results can not only reduce the computation complexity, but also highlight the importance of coordination to succeed in large groups.</div></div>","PeriodicalId":51063,"journal":{"name":"Information Sciences","volume":"708 ","pages":"Article 122070"},"PeriodicalIF":8.1,"publicationDate":"2025-03-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143579091","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}
Recommendation systems are essential tools for suggesting items or information to users based on their preferences and behaviors, which have been widely applied in various online platforms and services to personalize user experiences, increase user engagement, and drive business growth. However, the security and efficacy of recommendation systems can be compromised if the input data is tainted by malicious users. One of the primary threats to recommendation systems is shilling attacks, which pose great challenges in handling various types of huge-volume data with anomaly detection techniques. In this paper, we propose a novel anomaly detection framework named LTHiForest with the use of the learning to hash based isolation forest. Then, we instantiate the generic framework with one concrete hashing mechanism, extended order preserving hashing, to illustrate the stages of our framework and verify its effectiveness in detecting various anomalies. The core idea of this instantiation is to learn from data to construct a better isolation forest structure than the state-of-the-art methods like iForest and LSHiForest, which can achieve robust detection of various anomaly types. Extensive experiments on both synthetic and real-world data sets demonstrate the robustness and effectiveness of our framework for recommendation systems.
{"title":"A learning-based anomaly detection framework for secure recommendation","authors":"Haolong Xiang , Wenhao Fei , Ruiyang Ni , Xuyun Zhang","doi":"10.1016/j.ins.2025.122071","DOIUrl":"10.1016/j.ins.2025.122071","url":null,"abstract":"<div><div>Recommendation systems are essential tools for suggesting items or information to users based on their preferences and behaviors, which have been widely applied in various online platforms and services to personalize user experiences, increase user engagement, and drive business growth. However, the security and efficacy of recommendation systems can be compromised if the input data is tainted by malicious users. One of the primary threats to recommendation systems is shilling attacks, which pose great challenges in handling various types of huge-volume data with anomaly detection techniques. In this paper, we propose a novel anomaly detection framework named LTHiForest with the use of the learning to hash based isolation forest. Then, we instantiate the generic framework with one concrete hashing mechanism, extended order preserving hashing, to illustrate the stages of our framework and verify its effectiveness in detecting various anomalies. The core idea of this instantiation is to learn from data to construct a better isolation forest structure than the state-of-the-art methods like iForest and LSHiForest, which can achieve robust detection of various anomaly types. Extensive experiments on both synthetic and real-world data sets demonstrate the robustness and effectiveness of our framework for recommendation systems.</div></div>","PeriodicalId":51063,"journal":{"name":"Information Sciences","volume":"708 ","pages":"Article 122071"},"PeriodicalIF":8.1,"publicationDate":"2025-03-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143579489","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 : 2025-03-06DOI: 10.1016/j.ins.2025.122051
Zhi Kong, Shudi Zhai, Lifu Wang, Ge Guo
Link prediction in complex networks involves forecasting unknown or future connections. Traditional methods often rely heavily on network topology information. However, in complex networks with significant attribute information (i.e., attributed networks), relying solely on topology information often leads to limited accuracy in predicting node connections. To address this issue, this study explores link prediction methods for weighted/unweighted and attributed/non-attributed networks. A novel node similarity is introduced, which comprehensively considers multiple factors. Based on structural information, attribute information, and weight information, a general link prediction framework is proposed for four different network types. This framework contains three core modules: a structural similarity module, an attribute similarity module, and a weighted similarity module. Using these modules, four global similarity measurements are defined for different network types. Taking weighted and attributed networks as an example, a link prediction algorithm is designed, and three key parameters are analyzed. To validate the effectiveness of the proposed algorithms, experiments are conducted on four types of real-world network datasets. The experimental results demonstrate that the proposed algorithms exhibit significant advantages in terms of prediction accuracy and robustness.
{"title":"A general link prediction method based on path node information and source node information","authors":"Zhi Kong, Shudi Zhai, Lifu Wang, Ge Guo","doi":"10.1016/j.ins.2025.122051","DOIUrl":"10.1016/j.ins.2025.122051","url":null,"abstract":"<div><div>Link prediction in complex networks involves forecasting unknown or future connections. Traditional methods often rely heavily on network topology information. However, in complex networks with significant attribute information (i.e., attributed networks), relying solely on topology information often leads to limited accuracy in predicting node connections. To address this issue, this study explores link prediction methods for weighted/unweighted and attributed/non-attributed networks. A novel node similarity is introduced, which comprehensively considers multiple factors. Based on structural information, attribute information, and weight information, a general link prediction framework is proposed for four different network types. This framework contains three core modules: a structural similarity module, an attribute similarity module, and a weighted similarity module. Using these modules, four global similarity measurements are defined for different network types. Taking weighted and attributed networks as an example, a link prediction algorithm is designed, and three key parameters are analyzed. To validate the effectiveness of the proposed algorithms, experiments are conducted on four types of real-world network datasets. The experimental results demonstrate that the proposed algorithms exhibit significant advantages in terms of prediction accuracy and robustness.</div></div>","PeriodicalId":51063,"journal":{"name":"Information Sciences","volume":"709 ","pages":"Article 122051"},"PeriodicalIF":8.1,"publicationDate":"2025-03-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143601014","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 : 2025-03-06DOI: 10.1016/j.ins.2025.122048
Chunmao Jiang, Yongpeng Wang
In heterogeneous computing environments, coscheduling of the graphics processing unit (GPU) and central processing unit (CPU) poses substantial challenges because of the diverse hardware architectures and dynamic workload patterns. To address this, we propose a novel hierarchical three-way decision fusion (H3WDF) strategy that integrates multigranularity workload predictions and adaptive scheduling policies. H3WDF employs a three-tier decision-making process, achieving global coordination through selective aggregation of localized decisions while establishing a balance between efficiency and quality of service. Results of experimental evaluation in a heterogeneous environment comprising several GPUs demonstrate the superior performance of H3WDF across multiple metrics. For “large language model” workloads, H3WDF achieves remarkable prediction accuracy both for short- and long-term forecasts. H3WDF's three-way decision mechanism effectively distributes workloads, balancing between batched executions for training tasks and immediate executions for inference workloads. Resource utilization exhibits significant improvements across all GPU types, with particularly strong performance in the case of high-end GPUs. Compared with the state-of-the-art baselines, H3WDF substantially reduces job completion times, enhances energy efficiency, and consistently maintains high fairness in resource allocation.
{"title":"Hierarchical three-way decision fusion for multigranularity GPU-CPU coscheduling in hybrid computing systems","authors":"Chunmao Jiang, Yongpeng Wang","doi":"10.1016/j.ins.2025.122048","DOIUrl":"10.1016/j.ins.2025.122048","url":null,"abstract":"<div><div>In heterogeneous computing environments, coscheduling of the graphics processing unit (GPU) and central processing unit (CPU) poses substantial challenges because of the diverse hardware architectures and dynamic workload patterns. To address this, we propose a novel hierarchical three-way decision fusion (H3WDF) strategy that integrates multigranularity workload predictions and adaptive scheduling policies. H3WDF employs a three-tier decision-making process, achieving global coordination through selective aggregation of localized decisions while establishing a balance between efficiency and quality of service. Results of experimental evaluation in a heterogeneous environment comprising several GPUs demonstrate the superior performance of H3WDF across multiple metrics. For “large language model” workloads, H3WDF achieves remarkable prediction accuracy both for short- and long-term forecasts. H3WDF's three-way decision mechanism effectively distributes workloads, balancing between batched executions for training tasks and immediate executions for inference workloads. Resource utilization exhibits significant improvements across all GPU types, with particularly strong performance in the case of high-end GPUs. Compared with the state-of-the-art baselines, H3WDF substantially reduces job completion times, enhances energy efficiency, and consistently maintains high fairness in resource allocation.</div></div>","PeriodicalId":51063,"journal":{"name":"Information Sciences","volume":"708 ","pages":"Article 122048"},"PeriodicalIF":8.1,"publicationDate":"2025-03-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143561980","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 : 2025-03-06DOI: 10.1016/j.ins.2025.122049
Yueshen Xu , Shaoyuan Zhang , Honghao Gao , Yuyu Yin , Jingzhao Hu , Rui Li
Web services have been prevalently applied in many software development scenarios such as the development of many applications in the cloud, mobile networks, and Web. But Web services usually suffer from the serious issue of single functionality; thus in recent years, compositions of Web services, i.e., mashups, have become a popular choice, and have brought significant convenience in providing more comprehensive functionalities. But the diversity and number of Web services are expanding dramatically, resulting in an intractable challenge: how to effectively recommend Web services for mashup development. Researchers have proposed several recommendation approaches, but existing solutions are primarily applicable in a one-shot paradigm, which may introduce biases and usually lack explainability. In real-world scenarios, developers usually need to incorporate new Web services to address emerging challenges, implying that the development paradigm could be interactive. Moreover, existing approaches are prone to produce mediocre accuracy. To solve these issues, in this paper, we develop an innovative Multimodal Features-based Unbiased (MMFU) service recommendation framework for interactive mashup development, which takes full advantage of the multimodal features involved in the development procedure. Our MMFU framework encompasses two separate models developed to learn deep features from both text and graph structural information, and contains a feature fusion mechanism. Extensive experiments were performed on two real-world datasets, and the results revealed that the MMFU framework outperforms the compared existing state-of-the-art approaches, and has high explainability and the ability to counteract biases.
{"title":"Explainable service recommendation for interactive mashup development counteracting biases","authors":"Yueshen Xu , Shaoyuan Zhang , Honghao Gao , Yuyu Yin , Jingzhao Hu , Rui Li","doi":"10.1016/j.ins.2025.122049","DOIUrl":"10.1016/j.ins.2025.122049","url":null,"abstract":"<div><div>Web services have been prevalently applied in many software development scenarios such as the development of many applications in the cloud, mobile networks, and Web. But Web services usually suffer from the serious issue of single functionality; thus in recent years, compositions of Web services, i.e., mashups, have become a popular choice, and have brought significant convenience in providing more comprehensive functionalities. But the diversity and number of Web services are expanding dramatically, resulting in an intractable challenge: how to effectively recommend Web services for mashup development. Researchers have proposed several recommendation approaches, but existing solutions are primarily applicable in a one-shot paradigm, which may introduce biases and usually lack explainability. In real-world scenarios, developers usually need to incorporate new Web services to address emerging challenges, implying that the development paradigm could be interactive. Moreover, existing approaches are prone to produce mediocre accuracy. To solve these issues, in this paper, we develop an innovative <em>Multimodal Features-based Unbiased (MMFU)</em> service recommendation framework for interactive mashup development, which takes full advantage of the multimodal features involved in the development procedure. Our MMFU framework encompasses two separate models developed to learn deep features from both text and graph structural information, and contains a feature fusion mechanism. Extensive experiments were performed on two real-world datasets, and the results revealed that the MMFU framework outperforms the compared existing state-of-the-art approaches, and has high explainability and the ability to counteract biases.</div></div>","PeriodicalId":51063,"journal":{"name":"Information Sciences","volume":"708 ","pages":"Article 122049"},"PeriodicalIF":8.1,"publicationDate":"2025-03-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143561983","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.2025.122068
Chen Huang , Yingjie Song , Hongjiang Ma , Xiangbing Zhou , Wu Deng
Large-scale optimization problems (LSOPs) in science and technology bring great challenges to the performance of algorithms. Although Competitive swarm optimizer (CSO) is an effective method, some shortcomings still exist when handling LSOPs, such as premature convergence. Therefore, a novel multiple level CSO with dual evaluation criteria and global optimization (DEGMCSO) is proposed to seek optimal solutions of LSOPs. In this paper, dual evaluation criteria are inserted into the multiple comparison process of the losers and winners to assist the algorithm retain more high quality particles with the potential. In addition to fitness values, adaptive selection weight fitness-distance is designed as the other criterion for selecting winners and losers according to different optimization problems. Meanwhile, a simple global optimal modification strategy is employed to get high quality global best solution. By CEC2010 and CEC2013 function suits, the results indicate DEGMCSO outperforms some popular algorithms. Finally, DEGMCSO is applied to feather selection problems of high dimension classification in the real world. The simulation results show that compared with the original CSO algorithm, DEGMCSO can find the solution of 16 functions on CEC2010 test function set which is obviously better than the CSO algorithm.
{"title":"A multiple level competitive swarm optimizer based on dual evaluation criteria and global optimization for large-scale optimization problem","authors":"Chen Huang , Yingjie Song , Hongjiang Ma , Xiangbing Zhou , Wu Deng","doi":"10.1016/j.ins.2025.122068","DOIUrl":"10.1016/j.ins.2025.122068","url":null,"abstract":"<div><div>Large-scale optimization problems (LSOPs) in science and technology bring great challenges to the performance of algorithms. Although Competitive swarm optimizer (CSO) is an effective method, some shortcomings still exist when handling LSOPs, such as premature convergence. Therefore, a novel multiple level CSO with dual evaluation criteria and global optimization (DEGMCSO) is proposed to seek optimal solutions of LSOPs. In this paper, dual evaluation criteria are inserted into the multiple comparison process of the losers and winners to assist the algorithm retain more high quality particles with the potential. In addition to fitness values, adaptive selection weight fitness-distance is designed as the other criterion for selecting winners and losers according to different optimization problems. Meanwhile, a simple global optimal modification strategy is employed to get high quality global best solution. By CEC2010 and CEC2013 function suits, the results indicate DEGMCSO outperforms some popular algorithms. Finally, DEGMCSO is applied to feather selection problems of high dimension classification in the real world. The simulation results show that compared with the original CSO algorithm, DEGMCSO can find the solution of 16 functions on CEC2010 test function set which is obviously better than the CSO algorithm.</div></div>","PeriodicalId":51063,"journal":{"name":"Information Sciences","volume":"708 ","pages":"Article 122068"},"PeriodicalIF":8.1,"publicationDate":"2025-03-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143579087","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.2025.122065
Yanshan Xiao , Linbin Chen , Bo Liu
Ordinal regression deals with the classification problems that the classes are ranked in order. The majority of existing ordinal regression approaches are designed for single-view data, and only a little work is done on multi-view ordinal regression. However, these multi-view ordinal regression works mainly concentrate on the consensus information between different views, while the complementary information that is critical in multi-view learning is not adequately considered in learning the ordinal regression classifier. In this paper, we put forward the multi-view ordinal regression model that incorporates feature augmentation and privileged information learning (MORFP). Firstly, distinguished from the existing multi-view ordinal regression approaches that mainly embody the consensus principle, MORFP introduces the concept of privileged information learning and implements both the consensus and complementarity principles. Based on the concept of privileged information learning, we treat one view as the privileged information of another view, so that different views can supply complementary information to enhance each other. Secondly, considering that the distributions of data in distinct views may be are greatly different, we map those views to a common subspace and augment this subspace by incorporating the original features of each view. By combining the original features in the views and projected features in the common subspace, the learned ordinal regression classifier is expected to have better discriminative ability than that learned on only the projected features or the original features. Lastly, we employ a heuristic framework to resolve the learning problem of MORFP, which trains the multi-view ordinal regression classifier and optimizes the projection matrices alternately. Numerical studies on real-life datasets have demonstrated that MORFP performs explicitly better than the existing multi-view ordinal regression approaches.
{"title":"Multi-view ordinal regression with feature augmentation and privileged information learning","authors":"Yanshan Xiao , Linbin Chen , Bo Liu","doi":"10.1016/j.ins.2025.122065","DOIUrl":"10.1016/j.ins.2025.122065","url":null,"abstract":"<div><div>Ordinal regression deals with the classification problems that the classes are ranked in order. The majority of existing ordinal regression approaches are designed for single-view data, and only a little work is done on multi-view ordinal regression. However, these multi-view ordinal regression works mainly concentrate on the consensus information between different views, while the complementary information that is critical in multi-view learning is not adequately considered in learning the ordinal regression classifier. In this paper, we put forward the multi-view ordinal regression model that incorporates feature augmentation and privileged information learning (MORFP). Firstly, distinguished from the existing multi-view ordinal regression approaches that mainly embody the consensus principle, MORFP introduces the concept of privileged information learning and implements both the consensus and complementarity principles. Based on the concept of privileged information learning, we treat one view as the privileged information of another view, so that different views can supply complementary information to enhance each other. Secondly, considering that the distributions of data in distinct views may be are greatly different, we map those views to a common subspace and augment this subspace by incorporating the original features of each view. By combining the original features in the views and projected features in the common subspace, the learned ordinal regression classifier is expected to have better discriminative ability than that learned on only the projected features or the original features. Lastly, we employ a heuristic framework to resolve the learning problem of MORFP, which trains the multi-view ordinal regression classifier and optimizes the projection matrices alternately. Numerical studies on real-life datasets have demonstrated that MORFP performs explicitly better than the existing multi-view ordinal regression approaches.</div></div>","PeriodicalId":51063,"journal":{"name":"Information Sciences","volume":"708 ","pages":"Article 122065"},"PeriodicalIF":8.1,"publicationDate":"2025-03-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143561979","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}
This paper proposes a fixed-time control method integrating backlash, friction, and unknown time-varying delay compensation to achieve precise load position tracking and speed synchronization in multi-motor systems. First, the novel tracking and synchronization control strategies are developed based on adaptive neural network (NN) dual sliding modes. The sliding mode surfaces are designed based on tracking and synchronization errors, respectively, and adaptive neural networks are employed to approximate unknown nonlinear functions. This approach ensures fixed-time convergence independent of the initial states of the system, with convergence time determined a priori and capable of ensuring satisfactory dynamic performance. Secondly, a new exponential Lyapunov-Krasovskii functional is constructed to compensate for the uncertainties of time-varying delays without requiring prior knowledge of the upper bound of delay nonlinearities. Finally, the effectiveness of the proposed approach is validated through simulation results.
{"title":"Fixed-time control of multi-motor nonlinear systems via adaptive neural network dual sliding mode","authors":"Wanjun Jing , Meng Li , Yong Chen , Zhangyong Chen","doi":"10.1016/j.ins.2025.122061","DOIUrl":"10.1016/j.ins.2025.122061","url":null,"abstract":"<div><div>This paper proposes a fixed-time control method integrating backlash, friction, and unknown time-varying delay compensation to achieve precise load position tracking and speed synchronization in multi-motor systems. First, the novel tracking and synchronization control strategies are developed based on adaptive neural network (NN) dual sliding modes. The sliding mode surfaces are designed based on tracking and synchronization errors, respectively, and adaptive neural networks are employed to approximate unknown nonlinear functions. This approach ensures fixed-time convergence independent of the initial states of the system, with convergence time determined a priori and capable of ensuring satisfactory dynamic performance. Secondly, a new exponential Lyapunov-Krasovskii functional is constructed to compensate for the uncertainties of time-varying delays without requiring prior knowledge of the upper bound of delay nonlinearities. Finally, the effectiveness of the proposed approach is validated through simulation results.</div></div>","PeriodicalId":51063,"journal":{"name":"Information Sciences","volume":"708 ","pages":"Article 122061"},"PeriodicalIF":8.1,"publicationDate":"2025-03-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143561982","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.2025.122047
Wenbin Qian , Junqi Li , Xinxin Cai , Jintao Huang , Weiping Ding
Partial label learning is a weakly supervised framework in which each training sample is associated with a set of candidate labels, but only one among them is the true label. Feature selection is a technique for enhancing the ability of learning models to generalize effectively. However, a challenging problem in feature selection for partial label learning is the impact of ambiguous candidate labels. To address this, this paper proposes a granular ball-based partial label feature selection method via fuzzy correlation and redundancy. Firstly, the paper utilizes granular ball computing to obtain two granular ball sets that respectively reflect the supervision information from candidate and non-candidate labels. The relative density between two granular ball sets is used to obtain labeling confidence which can identify the ground-truth labels. Then, a novel fuzzy entropy is defined by combining consistency in the granular ball with fuzzy information entropy. Additionally, fuzzy mutual information is derived by considering the fuzzy entropy and the fuzzy similarity constrained by granular ball radius. Fuzzy correlation and redundancy is measured by granular ball-based fuzzy mutual information. A heuristic search strategy is used to rank the features according to the principle of maximizing relevance and minimizing redundancy. Finally, experimental results on five real-world datasets and eight controlled UCI datasets show that the proposed method obtains superior performance than other compared methods.
{"title":"Granular ball-based partial label feature selection via fuzzy correlation and redundancy","authors":"Wenbin Qian , Junqi Li , Xinxin Cai , Jintao Huang , Weiping Ding","doi":"10.1016/j.ins.2025.122047","DOIUrl":"10.1016/j.ins.2025.122047","url":null,"abstract":"<div><div>Partial label learning is a weakly supervised framework in which each training sample is associated with a set of candidate labels, but only one among them is the true label. Feature selection is a technique for enhancing the ability of learning models to generalize effectively. However, a challenging problem in feature selection for partial label learning is the impact of ambiguous candidate labels. To address this, this paper proposes a granular ball-based partial label feature selection method via fuzzy correlation and redundancy. Firstly, the paper utilizes granular ball computing to obtain two granular ball sets that respectively reflect the supervision information from candidate and non-candidate labels. The relative density between two granular ball sets is used to obtain labeling confidence which can identify the ground-truth labels. Then, a novel fuzzy entropy is defined by combining consistency in the granular ball with fuzzy information entropy. Additionally, fuzzy mutual information is derived by considering the fuzzy entropy and the fuzzy similarity constrained by granular ball radius. Fuzzy correlation and redundancy is measured by granular ball-based fuzzy mutual information. A heuristic search strategy is used to rank the features according to the principle of maximizing relevance and minimizing redundancy. Finally, experimental results on five real-world datasets and eight controlled UCI datasets show that the proposed method obtains superior performance than other compared methods.</div></div>","PeriodicalId":51063,"journal":{"name":"Information Sciences","volume":"709 ","pages":"Article 122047"},"PeriodicalIF":8.1,"publicationDate":"2025-03-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143611345","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.2025.122060
Marcos Paulo Silva Gôlo, José Gilberto Barbosa de Medeiros Junior, Diego Furtado Silva, Ricardo Marcondes Marcacini
One-class learning (OCL) for graph neural networks (GNNs) comprises a set of techniques applied when real-world problems are modeled through graphs and have a single class of interest. These methods may employ a two-step strategy: first representing the graph and then classifying its nodes. End-to-end methods learn the node representations while classifying the nodes in OCL process. We highlight three main gaps in this literature: (i) non-customized representations for OCL; (ii) the lack of constraints on hypersphere learning; and (iii) the lack of interpretability. This paper presents One-cLass Graph Autoencoder (OLGA), a new OCL for GNN approach. OLGA is an end-to-end method that learns low-dimensional representations for nodes while encapsulating interest nodes through a proposed and new hypersphere loss function. Furthermore, OLGA combines this new hypersphere loss with the graph autoencoder reconstruction loss to improve model learning. The reconstruction loss is a constraint to the sole use of the hypersphere loss that can bias the model to encapsulate all nodes. Finally, our low-dimensional representation makes the OLGA interpretable since we can visualize the representation learning at each epoch. OLGA achieved state-of-the-art results and outperformed six other methods with statistical significance while maintaining the learning process interpretability with its low-dimensional representations.
{"title":"One-class graph autoencoder: A new end-to-end, low-dimensional, and interpretable approach for node classification","authors":"Marcos Paulo Silva Gôlo, José Gilberto Barbosa de Medeiros Junior, Diego Furtado Silva, Ricardo Marcondes Marcacini","doi":"10.1016/j.ins.2025.122060","DOIUrl":"10.1016/j.ins.2025.122060","url":null,"abstract":"<div><div>One-class learning (OCL) for graph neural networks (GNNs) comprises a set of techniques applied when real-world problems are modeled through graphs and have a single class of interest. These methods may employ a two-step strategy: first representing the graph and then classifying its nodes. End-to-end methods learn the node representations while classifying the nodes in OCL process. We highlight three main gaps in this literature: (i) non-customized representations for OCL; (ii) the lack of constraints on hypersphere learning; and (iii) the lack of interpretability. This paper presents <strong><u>O</u></strong>ne-c<strong><u>L</u></strong>ass <strong><u>G</u></strong>raph <strong><u>A</u></strong>utoencoder (OLGA), a new OCL for GNN approach. OLGA is an end-to-end method that learns low-dimensional representations for nodes while encapsulating interest nodes through a proposed and new hypersphere loss function. Furthermore, OLGA combines this new hypersphere loss with the graph autoencoder reconstruction loss to improve model learning. The reconstruction loss is a constraint to the sole use of the hypersphere loss that can bias the model to encapsulate all nodes. Finally, our low-dimensional representation makes the OLGA interpretable since we can visualize the representation learning at each epoch. OLGA achieved state-of-the-art results and outperformed six other methods with statistical significance while maintaining the learning process interpretability with its low-dimensional representations.</div></div>","PeriodicalId":51063,"journal":{"name":"Information Sciences","volume":"708 ","pages":"Article 122060"},"PeriodicalIF":8.1,"publicationDate":"2025-03-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143549451","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