Pub Date : 2024-04-17DOI: 10.1109/TSIPN.2024.3388951
Yao Zhou;Giorgio Battistelli;Luigi Chisci;Lin Gao;Gaiyou Li;Ping Wei
This paper deals with multi-object fusion in the presence of misbehaving sensor nodes, due to faults or adversarial attacks. In this setting, the main challenge is to identify and then remove messages coming from corrupted nodes. To this end, a three-step method is proposed, where the first step consists of choosing a reference density among the received ones on the basis of a minimum upper median divergence criterion. Then, thresholding on the divergence from the reference density is performed to derive a subset of densities to be fused. Finally, the remaining densities are fused following either the generalized covariance intersection (GCI) or minimum information loss (MIL) criterion. The implementation of the proposed method for resilient fusion of labeled multi-Bernoulli densities is also discussed. Finally, the performance of the proposed approach is assessed via simulation experiments on centralized and decentralized multi-target tracking case studies.
{"title":"Median-Based Resilient Multi-Object Fusion With Application to LMB Densities","authors":"Yao Zhou;Giorgio Battistelli;Luigi Chisci;Lin Gao;Gaiyou Li;Ping Wei","doi":"10.1109/TSIPN.2024.3388951","DOIUrl":"10.1109/TSIPN.2024.3388951","url":null,"abstract":"This paper deals with multi-object fusion in the presence of misbehaving sensor nodes, due to faults or adversarial attacks. In this setting, the main challenge is to identify and then remove messages coming from corrupted nodes. To this end, a three-step method is proposed, where the first step consists of choosing a reference density among the received ones on the basis of a minimum upper median divergence criterion. Then, thresholding on the divergence from the reference density is performed to derive a subset of densities to be fused. Finally, the remaining densities are fused following either the \u0000<italic>generalized covariance intersection</i>\u0000 (GCI) or \u0000<italic>minimum information loss</i>\u0000 (MIL) criterion. The implementation of the proposed method for resilient fusion of labeled multi-Bernoulli densities is also discussed. Finally, the performance of the proposed approach is assessed via simulation experiments on centralized and decentralized multi-target tracking case studies.","PeriodicalId":56268,"journal":{"name":"IEEE Transactions on Signal and Information Processing over Networks","volume":"10 ","pages":"473-486"},"PeriodicalIF":3.2,"publicationDate":"2024-04-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"140612732","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"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 : 2024-04-12DOI: 10.1109/TSIPN.2024.3375594
Kang Du;Yu Xiang
Causal inference from observational data following the restricted structural causal models (SCMs) framework hinges largely on the asymmetry between cause and effect from the data generating mechanisms, such as non-Gaussianity or non-linearity. This methodology can be adapted to stationary time series, yet inferring causal relationships from nonstationary time series remains a challenging task. In this work, we propose a new class of restricted SCM, via a time-varying filter and stationary noise, and exploit the asymmetry from nonstationarity for causal identification in both bivariate and network settings. We propose efficient procedures by leveraging powerful estimates of the bivariate evolutionary spectra for slowly varying processes. Various synthetic and real datasets that involve high-order and non-smooth filters are evaluated to demonstrate the effectiveness of our proposed methodology.
{"title":"Causal Inference From Slowly Varying Nonstationary Processes","authors":"Kang Du;Yu Xiang","doi":"10.1109/TSIPN.2024.3375594","DOIUrl":"10.1109/TSIPN.2024.3375594","url":null,"abstract":"Causal inference from observational data following the restricted structural causal models (SCMs) framework hinges largely on the asymmetry between cause and effect from the data generating mechanisms, such as non-Gaussianity or non-linearity. This methodology can be adapted to stationary time series, yet inferring causal relationships from nonstationary time series remains a challenging task. In this work, we propose a new class of restricted SCM, via a time-varying filter and stationary noise, and exploit the asymmetry from nonstationarity for causal identification in both bivariate and network settings. We propose efficient procedures by leveraging powerful estimates of the bivariate evolutionary spectra for slowly varying processes. Various synthetic and real datasets that involve high-order and non-smooth filters are evaluated to demonstrate the effectiveness of our proposed methodology.","PeriodicalId":56268,"journal":{"name":"IEEE Transactions on Signal and Information Processing over Networks","volume":"10 ","pages":"403-416"},"PeriodicalIF":3.2,"publicationDate":"2024-04-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"140590674","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
In this contribution, the bipartite flocking control problem of a set of autonomous mobile agents over cooperation-competition networks is investigated. Two kinds of asynchronous communication scenarios are considered, where each agent communicates with the neighbors only at certain time instants determined by its own clock, but not at other time instants. In addition, each agent adjusts the control input at all time instants in the first asynchronous scenario, and adjusts the control input only at its communication time instants in the second asynchronous scenario. Nonlinear positive and negative weight functions are designed to describe the effect of the distance between agents on the cooperation/competition degree in real interaction scenarios, where the farther (closer) the distance, the weaker (stronger) the cooperation/competition degree. With the help of signed graph theory and sub-stochastic matrix, the dynamic models under different asynchronous scenarios are analyzed, and the algebraic conditions for achieving bipartite flocking control are established separately. At last, the effectiveness of algebraic conditions is verified through numerical simulations.
{"title":"Multi-Agent Bipartite Flocking Control Over Cooperation-Competition Networks With Asynchronous Communications","authors":"Zhuangzhuang Ma;Lei Shi;Kai Chen;Jinliang Shao;Yuhua Cheng","doi":"10.1109/TSIPN.2024.3384817","DOIUrl":"10.1109/TSIPN.2024.3384817","url":null,"abstract":"In this contribution, the bipartite flocking control problem of a set of autonomous mobile agents over cooperation-competition networks is investigated. Two kinds of asynchronous communication scenarios are considered, where each agent communicates with the neighbors only at certain time instants determined by its own clock, but not at other time instants. In addition, each agent adjusts the control input at all time instants in the first asynchronous scenario, and adjusts the control input only at its communication time instants in the second asynchronous scenario. Nonlinear positive and negative weight functions are designed to describe the effect of the distance between agents on the cooperation/competition degree in real interaction scenarios, where the farther (closer) the distance, the weaker (stronger) the cooperation/competition degree. With the help of signed graph theory and sub-stochastic matrix, the dynamic models under different asynchronous scenarios are analyzed, and the algebraic conditions for achieving bipartite flocking control are established separately. At last, the effectiveness of algebraic conditions is verified through numerical simulations.","PeriodicalId":56268,"journal":{"name":"IEEE Transactions on Signal and Information Processing over Networks","volume":"10 ","pages":"460-472"},"PeriodicalIF":3.2,"publicationDate":"2024-04-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"140590750","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"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 : 2024-04-03DOI: 10.1109/TSIPN.2024.3384814
Chun Liu;Bin Jiang;Yang Li;Ron J. Patton
This study investigates the distributed fault-tolerant consensus issue of multi-agent systems subject to complicated abrupt and incipient time-varying actuator faults in physical hierarchy and aperiodic denial-of-service (DoS) attacks in networked hierarchy. Decentralized estimators are devised to estimate consecutive system states and actuator faults. A unified framework with an absolute local output-based closed-loop estimator in decentralized fault estimation design and a relative broadcasting state-based open-loop estimator in distributed event-triggered fault-tolerant consensus design is developed. Criteria of exponential consensus of the faulty multi-agent systems under DoS attacks are derived by virtue of average dwelling time and attack frequency technique. Simulations are outlined to confirm the efficacy of the proposed distributed fault-tolerant consensus control algorithm based on an event-triggered mechanism.
本研究探讨了多代理系统的分布式容错共识问题，该系统在物理层次结构中受到复杂的突发性和萌芽期时变致动器故障的影响，在网络层次结构中受到非周期性拒绝服务（DoS）攻击的影响。我们设计了分散估计器来估计连续的系统状态和执行器故障。在分散式故障估计设计中，开发了基于绝对本地输出的闭环估计器；在分布式事件触发容错共识设计中，开发了基于相对广播状态的开环估计器。通过平均停留时间和攻击频率技术，得出了在 DoS 攻击下故障多代理系统的指数共识标准。仿真结果证实了基于事件触发机制的分布式容错共识控制算法的有效性。
{"title":"Distributed Event-Triggered Fault-Tolerant Consensus Control of Multi-Agent Systems Under DoS Attacks","authors":"Chun Liu;Bin Jiang;Yang Li;Ron J. Patton","doi":"10.1109/TSIPN.2024.3384814","DOIUrl":"https://doi.org/10.1109/TSIPN.2024.3384814","url":null,"abstract":"This study investigates the distributed fault-tolerant consensus issue of multi-agent systems subject to complicated abrupt and incipient time-varying actuator faults in physical hierarchy and aperiodic denial-of-service (DoS) attacks in networked hierarchy. Decentralized estimators are devised to estimate consecutive system states and actuator faults. A unified framework with an absolute local output-based closed-loop estimator in decentralized fault estimation design and a relative broadcasting state-based open-loop estimator in distributed event-triggered fault-tolerant consensus design is developed. Criteria of exponential consensus of the faulty multi-agent systems under DoS attacks are derived by virtue of average dwelling time and attack frequency technique. Simulations are outlined to confirm the efficacy of the proposed distributed fault-tolerant consensus control algorithm based on an event-triggered mechanism.","PeriodicalId":56268,"journal":{"name":"IEEE Transactions on Signal and Information Processing over Networks","volume":"10 ","pages":"390-402"},"PeriodicalIF":3.2,"publicationDate":"2024-04-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"140550146","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"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 : 2024-03-27DOI: 10.1109/TSIPN.2024.3382427
Pan Yu;Yifan Ding;Kang-Zhi Liu;Xiaoli Li
This paper develops a composite output consensus control protocol for a general linear multiagent system subject to mismatched disturbances, which incorporates active disturbance-rejection control and fully distributed adaptive consensus control. To estimate and then cancel out the effect of mismatched disturbances on the outputs of the agents, heterogeneous generalized equivalent-input-disturbance estimators are constructed in the inner loop. Then a fully distributed adaptive feedback controller is designed to achieve consensus control based on the states of the designed heterogeneous observers for the agents. The restriction on the disturbances is lowered, the requirement for the global information of the communication topology is removed, and the exchanging information among agents is only relative estimated states. Further, the output consensus performance is analyzed for the closed-loop multiagent system. Our results complement and improve the results of the existing literature. Lastly, the effectiveness and superiority of the developed method are demonstrated through a numerical simulation and a comparison with the distributed extended-state-observer-based method.
{"title":"Composite Output Consensus Control for General Linear Multiagent Systems With Heterogeneous Mismatched Disturbances","authors":"Pan Yu;Yifan Ding;Kang-Zhi Liu;Xiaoli Li","doi":"10.1109/TSIPN.2024.3382427","DOIUrl":"10.1109/TSIPN.2024.3382427","url":null,"abstract":"This paper develops a composite output consensus control protocol for a general linear multiagent system subject to mismatched disturbances, which incorporates active disturbance-rejection control and fully distributed adaptive consensus control. To estimate and then cancel out the effect of mismatched disturbances on the outputs of the agents, heterogeneous generalized equivalent-input-disturbance estimators are constructed in the inner loop. Then a fully distributed adaptive feedback controller is designed to achieve consensus control based on the states of the designed heterogeneous observers for the agents. The restriction on the disturbances is lowered, the requirement for the global information of the communication topology is removed, and the exchanging information among agents is only relative estimated states. Further, the output consensus performance is analyzed for the closed-loop multiagent system. Our results complement and improve the results of the existing literature. Lastly, the effectiveness and superiority of the developed method are demonstrated through a numerical simulation and a comparison with the distributed extended-state-observer-based method.","PeriodicalId":56268,"journal":{"name":"IEEE Transactions on Signal and Information Processing over Networks","volume":"10 ","pages":"434-444"},"PeriodicalIF":3.2,"publicationDate":"2024-03-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"140313490","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Graphs are a powerful tool for representing entities and their relationships. Current advances in graph signal processing have made it possible to analyze graph-based data more effectively. Recent research show that, to ensure critical sampling, manyfilterbank design algorithms are only applicable to bipartite graphs. However, general graph signals may not exist on a bipartite graph structure. To overcome this difficulty, we propose in this paper a novel algorithm to find a bipartite approximation to the original non-bipartite graph while preserving its global structure. To achieve this goal, the original bipartite graph approximation (BGA) problem is constructed based on eigenvalue optimization of adjacency matrix, which is then relaxed so as to obtain a closed-form solution. We introduce the alternating direction method of multipliers (ADMM) to achieve a single bipartite graph or a set of edge-disjoint bipartite subgraphs that approximates the original graph. Additionally, we develop a distributed version of the BGA to address the computational challenges when processing large-scale graphs. Experimental results demonstrate the effectiveness of the proposed method and suggest it as a promising alternative approach for bipartite graph decomposition.
{"title":"Bipartite Graph Approximation by Eigenvalue Optimization","authors":"Aimin Jiang;Xintong Shi;Yibin Tang;Yanping Zhu;Hon Keung Kwan","doi":"10.1109/TSIPN.2024.3380351","DOIUrl":"10.1109/TSIPN.2024.3380351","url":null,"abstract":"Graphs are a powerful tool for representing entities and their relationships. Current advances in graph signal processing have made it possible to analyze graph-based data more effectively. Recent research show that, to ensure critical sampling, manyfilterbank design algorithms are only applicable to bipartite graphs. However, general graph signals may not exist on a bipartite graph structure. To overcome this difficulty, we propose in this paper a novel algorithm to find a bipartite approximation to the original non-bipartite graph while preserving its global structure. To achieve this goal, the original bipartite graph approximation (BGA) problem is constructed based on eigenvalue optimization of adjacency matrix, which is then relaxed so as to obtain a closed-form solution. We introduce the alternating direction method of multipliers (ADMM) to achieve a single bipartite graph or a set of edge-disjoint bipartite subgraphs that approximates the original graph. Additionally, we develop a distributed version of the BGA to address the computational challenges when processing large-scale graphs. Experimental results demonstrate the effectiveness of the proposed method and suggest it as a promising alternative approach for bipartite graph decomposition.","PeriodicalId":56268,"journal":{"name":"IEEE Transactions on Signal and Information Processing over Networks","volume":"10 ","pages":"307-319"},"PeriodicalIF":3.2,"publicationDate":"2024-03-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"140299077","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
The problem of distributed optimization requires a group of networked agents to compute a parameter that minimizes the average of their local cost functions. While there are a variety of distributed optimization algorithms that can solve this problem, they are typically vulnerable to “Byzantine” agents that do not follow the algorithm. Recent attempts to address this issue focus on single dimensional functions, or assume certain statistical properties of the functions at the agents. In this paper, we provide two resilient, scalable, distributed optimization algorithms for multi-dimensional functions. Our schemes involve two filters, (1) a distance-based filter and (2) a min-max filter, which each remove neighborhood states that are extreme (defined precisely in our algorithms) at each iteration. We show that these algorithms can mitigate the impact of up to $F$ (unknown) Byzantine agents in the neighborhood of each regular agent. In particular, we show that if the network topology satisfies certain conditions, all of the regular agents' states are guaranteed to converge to a bounded region that contains the minimizer of the average of the regular agents' functions.
{"title":"Scalable Distributed Optimization of Multi-Dimensional Functions Despite Byzantine Adversaries","authors":"Kananart Kuwaranancharoen;Lei Xin;Shreyas Sundaram","doi":"10.1109/TSIPN.2024.3379844","DOIUrl":"10.1109/TSIPN.2024.3379844","url":null,"abstract":"The problem of distributed optimization requires a group of networked agents to compute a parameter that minimizes the average of their local cost functions. While there are a variety of distributed optimization algorithms that can solve this problem, they are typically vulnerable to “Byzantine” agents that do not follow the algorithm. Recent attempts to address this issue focus on single dimensional functions, or assume certain statistical properties of the functions at the agents. In this paper, we provide two resilient, scalable, distributed optimization algorithms for multi-dimensional functions. Our schemes involve two filters, (1) a distance-based filter and (2) a min-max filter, which each remove neighborhood states that are extreme (defined precisely in our algorithms) at each iteration. We show that these algorithms can mitigate the impact of up to \u0000<inline-formula><tex-math>$F$</tex-math></inline-formula>\u0000 (unknown) Byzantine agents in the neighborhood of each regular agent. In particular, we show that if the network topology satisfies certain conditions, all of the regular agents' states are guaranteed to converge to a bounded region that contains the minimizer of the average of the regular agents' functions.","PeriodicalId":56268,"journal":{"name":"IEEE Transactions on Signal and Information Processing over Networks","volume":"10 ","pages":"360-375"},"PeriodicalIF":3.2,"publicationDate":"2024-03-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"140196592","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"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 : 2024-03-21DOI: 10.1109/TSIPN.2024.3380362
Arda Can Aras;Tuna Alikaşifoğlu;Aykut Koç
By employing attention mechanisms, transformers have made great improvements in nearly all NLP tasks, including text classification. However, the context of the transformer's attention mechanism is limited to single sequences, and their fine-tuning stage can utilize only inductive learning. Focusing on broader contexts by representing texts as graphs, previous works have generalized transformer models to graph domains to employ attention mechanisms beyond single sequences. However, these approaches either require exhaustive pre-training stages, learn only transductively, or can learn inductively without utilizing pre-trained models. To address these problems simultaneously, we propose the Graph Receptive Transformer Encoder (GRTE), which combines graph neural networks (GNNs) with large-scale pre-trained models for text classification in both inductive and transductive fashions. By constructing heterogeneous and homogeneous graphs over given corpora and not requiring a pre-training stage, GRTE can utilize information from both large-scale pre-trained models and graph-structured relations. Our proposed method retrieves global and contextual information in documents and generates word embeddings as a by-product of inductive inference. We compared the proposed GRTE with a wide range of baseline models through comprehensive experiments. Compared to the state-of-the-art, we demonstrated that GRTE improves model performances and offers computational savings up to ˜100×.
{"title":"Graph Receptive Transformer Encoder for Text Classification","authors":"Arda Can Aras;Tuna Alikaşifoğlu;Aykut Koç","doi":"10.1109/TSIPN.2024.3380362","DOIUrl":"10.1109/TSIPN.2024.3380362","url":null,"abstract":"By employing attention mechanisms, transformers have made great improvements in nearly all NLP tasks, including text classification. However, the context of the transformer's attention mechanism is limited to single sequences, and their fine-tuning stage can utilize only inductive learning. Focusing on broader contexts by representing texts as graphs, previous works have generalized transformer models to graph domains to employ attention mechanisms beyond single sequences. However, these approaches either require exhaustive pre-training stages, learn only transductively, or can learn inductively without utilizing pre-trained models. To address these problems simultaneously, we propose the Graph Receptive Transformer Encoder (GRTE), which combines graph neural networks (GNNs) with large-scale pre-trained models for text classification in both inductive and transductive fashions. By constructing heterogeneous and homogeneous graphs over given corpora and not requiring a pre-training stage, GRTE can utilize information from both large-scale pre-trained models and graph-structured relations. Our proposed method retrieves global and contextual information in documents and generates word embeddings as a by-product of inductive inference. We compared the proposed GRTE with a wide range of baseline models through comprehensive experiments. Compared to the state-of-the-art, we demonstrated that GRTE improves model performances and offers computational savings up to ˜100×.","PeriodicalId":56268,"journal":{"name":"IEEE Transactions on Signal and Information Processing over Networks","volume":"10 ","pages":"347-359"},"PeriodicalIF":3.2,"publicationDate":"2024-03-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"140205723","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"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 : 2024-03-21DOI: 10.1109/TSIPN.2024.3380374
Xiaoxian Lao;Chunguang Li
Distributed estimation over networks has drawn much attention in recent years. In the problem of distributed estimation, a set of nodes is requested to estimate some parameter of interest from noisy measurements. The nodes interact with each other to carry out the task jointly. Many algorithms have been proposed for solving the distributed estimation problem, among which the diffusion strategy is well-accepted. Information diffusion among nodes consumes bandwidth and energy resources, while in real-world applications these resources are limited. To cope with the resources constraint, partial diffusion schemes are developed. Each node only disseminates a subset of entries of interested vector in each interaction. Besides the partial transmission, quantization is another widely adopted technique for saving the communication resources. The two methods work in different aspects and can be considered jointly to make the communication more efficient. In this paper, we propose a partial diffusion scheme with quantization. An optimization problem for communication resources allocation is formulated and solved. In each interaction, the nodes will adaptively determine whether to transmit more entries or assign more bits to quantize each entry. We derive sufficient conditions for convergence of the overall algorithm. We also demonstrate the advantages of the proposed scheme in terms of both convergence speed and estimation accuracy.
{"title":"Partial Diffusion With Quantization Over Networks","authors":"Xiaoxian Lao;Chunguang Li","doi":"10.1109/TSIPN.2024.3380374","DOIUrl":"10.1109/TSIPN.2024.3380374","url":null,"abstract":"Distributed estimation over networks has drawn much attention in recent years. In the problem of distributed estimation, a set of nodes is requested to estimate some parameter of interest from noisy measurements. The nodes interact with each other to carry out the task jointly. Many algorithms have been proposed for solving the distributed estimation problem, among which the diffusion strategy is well-accepted. Information diffusion among nodes consumes bandwidth and energy resources, while in real-world applications these resources are limited. To cope with the resources constraint, partial diffusion schemes are developed. Each node only disseminates a subset of entries of interested vector in each interaction. Besides the partial transmission, quantization is another widely adopted technique for saving the communication resources. The two methods work in different aspects and can be considered jointly to make the communication more efficient. In this paper, we propose a partial diffusion scheme with quantization. An optimization problem for communication resources allocation is formulated and solved. In each interaction, the nodes will adaptively determine whether to transmit more entries or assign more bits to quantize each entry. We derive sufficient conditions for convergence of the overall algorithm. We also demonstrate the advantages of the proposed scheme in terms of both convergence speed and estimation accuracy.","PeriodicalId":56268,"journal":{"name":"IEEE Transactions on Signal and Information Processing over Networks","volume":"10 ","pages":"320-331"},"PeriodicalIF":3.2,"publicationDate":"2024-03-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"140196492","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Distributed optimization is manifesting great potential in multiple fields, e.g., machine learning, control, resource allocation, etc. Existing decentralized optimization algorithms require sharing explicit state information among the agents, which raises the risk of private information leakage. To ensure privacy security, combining information security mechanisms, such as differential privacy and homomorphic encryption, with traditional decentralized optimization algorithms is a commonly used means. However, this may either sacrifice optimization accuracy or incur a heavy computational burden. To overcome these shortcomings, we develop a novel privacy-preserving decentralized optimization algorithm, named PPSD, that combines gradient tracking with a state decomposition mechanism. Specifically, each agent decomposes its state associated with the gradient into two substates. One substate is used for interaction with neighboring agents, and the other substate containing private information acts only on the first substate and thus is entirely agnostic to other agents. When the objective function is smooth and satisfies the Polyak-Łojasiewicz (PL) condition, PPSD attains an $R$ -linear convergence rate. Moreover, the algorithm can preserve the normal agents' private information from being leaked to honest-but-curious attackers. Simulations further confirm the results.
{"title":"Privacy-Preserving Push-Pull Method for Decentralized Optimization via State Decomposition","authors":"Huqiang Cheng;Xiaofeng Liao;Huaqing Li;Qingguo Lü;You Zhao","doi":"10.1109/TSIPN.2024.3402430","DOIUrl":"10.1109/TSIPN.2024.3402430","url":null,"abstract":"Distributed optimization is manifesting great potential in multiple fields, e.g., machine learning, control, resource allocation, etc. Existing decentralized optimization algorithms require sharing explicit state information among the agents, which raises the risk of private information leakage. To ensure privacy security, combining information security mechanisms, such as differential privacy and homomorphic encryption, with traditional decentralized optimization algorithms is a commonly used means. However, this may either sacrifice optimization accuracy or incur a heavy computational burden. To overcome these shortcomings, we develop a novel privacy-preserving decentralized optimization algorithm, named PPSD, that combines gradient tracking with a state decomposition mechanism. Specifically, each agent decomposes its state associated with the gradient into two substates. One substate is used for interaction with neighboring agents, and the other substate containing private information acts only on the first substate and thus is entirely agnostic to other agents. When the objective function is smooth and satisfies the Polyak-Łojasiewicz (PL) condition, PPSD attains an \u0000<inline-formula><tex-math>$R$</tex-math></inline-formula>\u0000-linear convergence rate. Moreover, the algorithm can preserve the normal agents' private information from being leaked to honest-but-curious attackers. Simulations further confirm the results.","PeriodicalId":56268,"journal":{"name":"IEEE Transactions on Signal and Information Processing over Networks","volume":"10 ","pages":"513-526"},"PeriodicalIF":3.2,"publicationDate":"2024-03-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141151780","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：