Stochastic variational inference is an efficient Bayesian inference technology for massive datasets,which approximates posteriors by using noisy gradient estimates.Traditional stochastic vari-ational inference can only be performed in a centralized manner,which limits its applications in a wide range of situations where data is possessed by multiple nodes.Therefore,this paper develops a novel trust-region based stochastic variational inference algorithm for a general class of conjugate-exponential models over distributed and asynchronous networks,where the global parameters are diffused over the network by using the Metropolis rule and the local parameters are updated by using the trust-region method.Besides,a simple rule is introduced to balance the transmission frequencies between neighbor-ing nodes such that the proposed distributed algorithm can be performed in an asynchronous manner.The utility of the proposed algorithm is tested by fitting the Bernoulli model and the Gaussian model to different datasets on a synthetic network,and experimental results demonstrate its effectiveness and advantages over existing works.

FU Weiming;QIN Jiahu;LING Qing;KANG Yu;YE Baijia

Department of Automation,University of Science and Technology of China,Hefei 230027,China;Institute of Artificial Intelligence, Hefei Comprehensive National Science Center, Hefei 230088, China;School of Computer Science and Engineering,and Guangdong Province Key Laboratory of Computational Sci-ence,Sun Yat-Sen University,Guanazhou 510006,China;Institute of Advanced Technology, University of Science and Technology of China, Hefei 230027, China

系统科学与复杂性学报（英文版）

[1]FU Weiming;QIN Jiahu;LING Qing;KANG Yu;YE Baijia-.Trust-Region Based Stochastic Variational Inference for Distributed and Asynchronous Networks)[J].系统科学与复杂性学报（英文版）,2022(06):2062-2076

posteriors

Trust,Region,Based,Stochastic,Variational,Inference,Distributed,Asynchronous,Networks,variational,inference,efficient,Bayesian,technology,massive,datasets,which,approximates,by,using,noisy,gradient,estimates,Traditional,stochastic,can,only,performed,centralized,manner,limits,applications,wide,range,situations,where,possessed,multiple,nodes,Therefore,this,paper,develops,novel,trust,region,algorithm,general,class,conjugate,exponential,models,over,distributed,asynchronous,networks,global,parameters,are,diffused,Metropolis,rule,local,updated,method,Besides,simple,introduced,balance,transmission,frequencies,between,neighbor,such,that,proposed,utility,tested,fitting,Bernoulli,Gaussian,different,synthetic,experimental,results,demonstrate,effectiveness,advantages,existing

0.596446