长尾分类在现实世界中是一项不可避免且充满挑战的任务。传统方法通常只专注于类间的不平衡分布,然而近期的研究开始重视类内的长尾分布,即同一类别内,具有头部属性的样本远多于尾部属性的样本。由于属性的隐含性和其组合的复杂性,类内...长尾分类在现实世界中是一项不可避免且充满挑战的任务。传统方法通常只专注于类间的不平衡分布,然而近期的研究开始重视类内的长尾分布,即同一类别内,具有头部属性的样本远多于尾部属性的样本。由于属性的隐含性和其组合的复杂性,类内不平衡问题更加难以处理。为此,文中提出一种基于引领森林并使用多中心损失的广义长尾分类框架(Cognisance),旨在通过不变性特征学习的范式建立长尾分类问题的多粒度联合求解模型。首先,该框架通过无监督学习构建粗粒度引领森林(Coarse-Grained Leading Forest,CLF),以更好地表征类内关于不同属性的样本分布,进而在不变风险最小化的过程中构建不同的环境。其次,设计了一种新的度量学习损失,即多中心损失(Multi-Center Loss,MCL),可在特征学习过程中逐步消除混淆属性。同时,Cognisance不依赖于特定模型结构,可作为独立组件与其他长尾分类方法集成。在ImageNet-GLT和MSCOCO-GLT数据集上的实验结果显示,所提框架取得了最佳性能,现有方法通过与本框架集成,在Top1-Accuracy指标上均获得2%~8%的提升。展开更多
This paper discusses problems of multicasting and its application. Two ways are proposed to overcome theproblems of multicasting according to the architectures of current networks and the functional requirements of as...This paper discusses problems of multicasting and its application. Two ways are proposed to overcome theproblems of multicasting according to the architectures of current networks and the functional requirements of asurveillance System. One is that Winsock API is directly used to implement multicasting on a network supportingmulticasting. The other adopts the technology of proxy server and tunneling on a network not supporting multicast-ing. In addition,this paper discusses some techniques of multicasting,such as sending,resending,auto-adjusting withthe change of the transmission quality of networks,and the management of multicast group and multicast addresses.展开更多
Multi-Agent Collaborative Problem Solving is one basic issue of the research of Multi-Agent Systen(MAS). In this paper we summarize some research work of Multi-Agent collaborative problem solving,expound thecharacteri...Multi-Agent Collaborative Problem Solving is one basic issue of the research of Multi-Agent Systen(MAS). In this paper we summarize some research work of Multi-Agent collaborative problem solving,expound thecharacteristic of Multi-Agnet collaborative problem solving,Model of Multi-Agent collaborative problem solving,pro-cess of solving、the application field of Multi-Agent collaborative problem solving and some challenge. Especially wediscuss the main models ,introduce the representative model including joint-intention,joint-commitment ,shared plan.展开更多
文摘长尾分类在现实世界中是一项不可避免且充满挑战的任务。传统方法通常只专注于类间的不平衡分布,然而近期的研究开始重视类内的长尾分布,即同一类别内,具有头部属性的样本远多于尾部属性的样本。由于属性的隐含性和其组合的复杂性,类内不平衡问题更加难以处理。为此,文中提出一种基于引领森林并使用多中心损失的广义长尾分类框架(Cognisance),旨在通过不变性特征学习的范式建立长尾分类问题的多粒度联合求解模型。首先,该框架通过无监督学习构建粗粒度引领森林(Coarse-Grained Leading Forest,CLF),以更好地表征类内关于不同属性的样本分布,进而在不变风险最小化的过程中构建不同的环境。其次,设计了一种新的度量学习损失,即多中心损失(Multi-Center Loss,MCL),可在特征学习过程中逐步消除混淆属性。同时,Cognisance不依赖于特定模型结构,可作为独立组件与其他长尾分类方法集成。在ImageNet-GLT和MSCOCO-GLT数据集上的实验结果显示,所提框架取得了最佳性能,现有方法通过与本框架集成,在Top1-Accuracy指标上均获得2%~8%的提升。
文摘This paper discusses problems of multicasting and its application. Two ways are proposed to overcome theproblems of multicasting according to the architectures of current networks and the functional requirements of asurveillance System. One is that Winsock API is directly used to implement multicasting on a network supportingmulticasting. The other adopts the technology of proxy server and tunneling on a network not supporting multicast-ing. In addition,this paper discusses some techniques of multicasting,such as sending,resending,auto-adjusting withthe change of the transmission quality of networks,and the management of multicast group and multicast addresses.
文摘Multi-Agent Collaborative Problem Solving is one basic issue of the research of Multi-Agent Systen(MAS). In this paper we summarize some research work of Multi-Agent collaborative problem solving,expound thecharacteristic of Multi-Agnet collaborative problem solving,Model of Multi-Agent collaborative problem solving,pro-cess of solving、the application field of Multi-Agent collaborative problem solving and some challenge. Especially wediscuss the main models ,introduce the representative model including joint-intention,joint-commitment ,shared plan.
