In order to improve the efficiency and quality of service composition,a service composition algorithm based on semantic constraint is proposed.First, a user’s requirements and services from a service repository are c...In order to improve the efficiency and quality of service composition,a service composition algorithm based on semantic constraint is proposed.First, a user’s requirements and services from a service repository are compared with the help of a matching algorithm.The algorithm has two levels and filters out the services which do not match the user’s constraint personality requirements.The mechanism can reduce the searching scope at the beginning of the service composition algorithm.Secondly,satisfactions of those selected services for the user’s personality requirements are computed and those services,which have the greatest satisfaction value to make up the service composition,are used.The algorithm is evaluated analytically and experimentally based on the efficiency of service composition and satisfaction for the user’s personality requirements.展开更多
传统联邦学习中经过加权聚合得到的全局模型无法应对跨客户端的数据异构的问题。现有研究通过形成个性化模型应对,但个性化模型如何平衡全局的共性信息和本地的个性信息是一个挑战。针对上述问题,提出了一种个性化联邦学习模型聚合框架F...传统联邦学习中经过加权聚合得到的全局模型无法应对跨客户端的数据异构的问题。现有研究通过形成个性化模型应对,但个性化模型如何平衡全局的共性信息和本地的个性信息是一个挑战。针对上述问题,提出了一种个性化联邦学习模型聚合框架FedPG(federated learning with personalized global model)。FedPG基于客户端模型的相似性,将归一化后的模型参数变化量的余弦相似度作为模型聚合的个性化权重,从而实现面向客户端的全局模型个性化聚合。通过引入平滑系数,该框架可以灵活地调整模型中共性信息和个性信息的比重。为了降低平滑系数的选择成本,进一步提出调度平滑系数的个性化联邦学习模型聚合框架FedPGS(federated learning with personalized global model and scheduled personalization)。在实验中,FedPG和FedPGS两个框架使得FedAvg、FedProto、FedProx算法在特征分布偏移的数据集上的准确率平均提升1.20~11.50百分点,且使得模型的准确率受恶意设备的影响更小。结果表明,FedPG和FedPGS框架在数据异构和存在恶意设备干扰的情况下能有效提升模型的准确率和鲁棒性。展开更多
A personalized recommendation for cloud services, which is based on usage history and the cooperative relationship of cloud services, is presented. According to service groups, a service group could be defined as seve...A personalized recommendation for cloud services, which is based on usage history and the cooperative relationship of cloud services, is presented. According to service groups, a service group could be defined as several services that were used together by one user at a time, and cooperative relationship between each two services can be calculated. In the process of recommendation, the services which are highly related to the service that the user has selected would be obtained firstly, the result should then take the QoS (Quality of Service) similarity between service’s QoS and user’s preference into account, so the final result combining the cooperative relationship and similarity will meet the functional needs of users and also meet the user’s personalized non-functional requirements. The simulation proves that the algorithm works effectively.展开更多
In person re-IDentification (re-ID) task,the learning of part-level features benefits from fine-grained information.To facilitate part alignment,which is a prerequisite for learning part-level features,a popular appro...In person re-IDentification (re-ID) task,the learning of part-level features benefits from fine-grained information.To facilitate part alignment,which is a prerequisite for learning part-level features,a popular approach is to detect semantic parts with the use of human parsing or pose estimation.Such methods of semantic partition do offer cues to good part alignment but are prone to noisy part detection,especially when they are employed in an off-the-shelf manner.In response,this paper proposes a novel part feature learning method for re-ID,that suppresses the impact of noisy semantic part detection through Supervised Non-local Similarity (SNS) learning.Given several detected semantic parts,SNS first locates their center points on the convolutional feature maps for use as a set of anchors and then evaluates the similarity values between these anchors and each pixel on the feature maps.The non-local similarity learning is supervised such that:each anchor should be similar to itself and simultaneously dissimilar to any other anchors,thus yielding the SNS.Finally,each anchor absorbs features from all of the similar pixels on the convolutional feature maps to generate a corresponding part feature (SNS feature).We evaluate our method with extensive experiments conducted under both holistic and partial re-ID scenarios.Experimental results confirm that SNS consistently improves re-ID accuracy using human parsing or pose estimation,and that our results are on par with state-of-the-art methods.展开更多
基金The National Natural Science Foundation of China(No.60673130)the Natural Science Foundation of Shandong Province(No.Y2006G29,Y2007G24,Y2007G38)
文摘In order to improve the efficiency and quality of service composition,a service composition algorithm based on semantic constraint is proposed.First, a user’s requirements and services from a service repository are compared with the help of a matching algorithm.The algorithm has two levels and filters out the services which do not match the user’s constraint personality requirements.The mechanism can reduce the searching scope at the beginning of the service composition algorithm.Secondly,satisfactions of those selected services for the user’s personality requirements are computed and those services,which have the greatest satisfaction value to make up the service composition,are used.The algorithm is evaluated analytically and experimentally based on the efficiency of service composition and satisfaction for the user’s personality requirements.
文摘传统联邦学习中经过加权聚合得到的全局模型无法应对跨客户端的数据异构的问题。现有研究通过形成个性化模型应对,但个性化模型如何平衡全局的共性信息和本地的个性信息是一个挑战。针对上述问题,提出了一种个性化联邦学习模型聚合框架FedPG(federated learning with personalized global model)。FedPG基于客户端模型的相似性,将归一化后的模型参数变化量的余弦相似度作为模型聚合的个性化权重,从而实现面向客户端的全局模型个性化聚合。通过引入平滑系数,该框架可以灵活地调整模型中共性信息和个性信息的比重。为了降低平滑系数的选择成本,进一步提出调度平滑系数的个性化联邦学习模型聚合框架FedPGS(federated learning with personalized global model and scheduled personalization)。在实验中,FedPG和FedPGS两个框架使得FedAvg、FedProto、FedProx算法在特征分布偏移的数据集上的准确率平均提升1.20~11.50百分点,且使得模型的准确率受恶意设备的影响更小。结果表明,FedPG和FedPGS框架在数据异构和存在恶意设备干扰的情况下能有效提升模型的准确率和鲁棒性。
文摘A personalized recommendation for cloud services, which is based on usage history and the cooperative relationship of cloud services, is presented. According to service groups, a service group could be defined as several services that were used together by one user at a time, and cooperative relationship between each two services can be calculated. In the process of recommendation, the services which are highly related to the service that the user has selected would be obtained firstly, the result should then take the QoS (Quality of Service) similarity between service’s QoS and user’s preference into account, so the final result combining the cooperative relationship and similarity will meet the functional needs of users and also meet the user’s personalized non-functional requirements. The simulation proves that the algorithm works effectively.
基金supported by the National Key Research and Development Program of China(No.2016YFB0801301)the National Natural Science Foundation of China(No.61771288)。
文摘In person re-IDentification (re-ID) task,the learning of part-level features benefits from fine-grained information.To facilitate part alignment,which is a prerequisite for learning part-level features,a popular approach is to detect semantic parts with the use of human parsing or pose estimation.Such methods of semantic partition do offer cues to good part alignment but are prone to noisy part detection,especially when they are employed in an off-the-shelf manner.In response,this paper proposes a novel part feature learning method for re-ID,that suppresses the impact of noisy semantic part detection through Supervised Non-local Similarity (SNS) learning.Given several detected semantic parts,SNS first locates their center points on the convolutional feature maps for use as a set of anchors and then evaluates the similarity values between these anchors and each pixel on the feature maps.The non-local similarity learning is supervised such that:each anchor should be similar to itself and simultaneously dissimilar to any other anchors,thus yielding the SNS.Finally,each anchor absorbs features from all of the similar pixels on the convolutional feature maps to generate a corresponding part feature (SNS feature).We evaluate our method with extensive experiments conducted under both holistic and partial re-ID scenarios.Experimental results confirm that SNS consistently improves re-ID accuracy using human parsing or pose estimation,and that our results are on par with state-of-the-art methods.
基金国家高技术研究发展计划(863) (the National High- Tech Research and Development Plan of China under Grant No.2002AA414060)陕西省自然科学基金(the Natural Science Foundation of Shaanxi Province of China under Grant No.2005F05)
