Reforms in social governance are promoted by changes in social environments.Social governance models cannot be constant,and the rise of every innovative governance model is inseparable from an ever-changing society.Th...Reforms in social governance are promoted by changes in social environments.Social governance models cannot be constant,and the rise of every innovative governance model is inseparable from an ever-changing society.The theoretical connotation of the collaborative innovation model in regional social governance is mainly reflected in its agents,actions,and processes.The fundamental requirements of the collaborative innovation model in regional social governance are de-administration and market-based.The key elements of the collaborative regional social governance model include:delegating power from central authorities,transferring selected social governance affairs,enhancing the development of governance capabilities in social governance agents,diversifying these agents by giving full play to the agency to strengthen the concept of collaborative governance for social governance agents,strengthening the establishment of the governance system,and ensuring the normal progress of the governance processes.The collaborative innovation model in regional social governance should be built in three aspects:a system of institutional norms,an evaluation indicator system,and law-based collaborative governance.展开更多
近几年,基于Transformer的预训练模型展现了强大的模态表征能力,促使了多模态的下游任务(如图像描述生成任务)正朝着完全端到端范式的趋势所转变,并且能够使得模型获得更好的性能以及更快的推理速度.然而,该技术所提取的网格型视觉特征...近几年,基于Transformer的预训练模型展现了强大的模态表征能力,促使了多模态的下游任务(如图像描述生成任务)正朝着完全端到端范式的趋势所转变,并且能够使得模型获得更好的性能以及更快的推理速度.然而,该技术所提取的网格型视觉特征中缺乏区域型的视觉信息,从而导致模型对对象内容的描述不精确.因此,预训练模型在图像描述生成任务上的适用性在很大程度上仍有待探索.针对这一问题,提出一种基于视觉区域聚合与双向协作学习的端到端图像描述生成方法(visual region aggregation and dual-level collaboration,VRADC).为了学习到区域型的视觉信息,设计了一种视觉区域聚合模块,将有相似语义的网格特征聚合在一起形成紧凑的视觉区域表征.接着,双向协作模块利用交叉注意力机制从两种视觉特征中学习到更加有代表性的语义信息,进而指导模型生成更加细粒度的图像描述文本.基于MSCOCO和Flickr30k两个数据集的实验结果表明,所提的VRADC方法能够大幅度地提升图像描述生成的质量,实现了最先进的性能.展开更多
文摘Reforms in social governance are promoted by changes in social environments.Social governance models cannot be constant,and the rise of every innovative governance model is inseparable from an ever-changing society.The theoretical connotation of the collaborative innovation model in regional social governance is mainly reflected in its agents,actions,and processes.The fundamental requirements of the collaborative innovation model in regional social governance are de-administration and market-based.The key elements of the collaborative regional social governance model include:delegating power from central authorities,transferring selected social governance affairs,enhancing the development of governance capabilities in social governance agents,diversifying these agents by giving full play to the agency to strengthen the concept of collaborative governance for social governance agents,strengthening the establishment of the governance system,and ensuring the normal progress of the governance processes.The collaborative innovation model in regional social governance should be built in three aspects:a system of institutional norms,an evaluation indicator system,and law-based collaborative governance.
文摘近几年,基于Transformer的预训练模型展现了强大的模态表征能力,促使了多模态的下游任务(如图像描述生成任务)正朝着完全端到端范式的趋势所转变,并且能够使得模型获得更好的性能以及更快的推理速度.然而,该技术所提取的网格型视觉特征中缺乏区域型的视觉信息,从而导致模型对对象内容的描述不精确.因此,预训练模型在图像描述生成任务上的适用性在很大程度上仍有待探索.针对这一问题,提出一种基于视觉区域聚合与双向协作学习的端到端图像描述生成方法(visual region aggregation and dual-level collaboration,VRADC).为了学习到区域型的视觉信息,设计了一种视觉区域聚合模块,将有相似语义的网格特征聚合在一起形成紧凑的视觉区域表征.接着,双向协作模块利用交叉注意力机制从两种视觉特征中学习到更加有代表性的语义信息,进而指导模型生成更加细粒度的图像描述文本.基于MSCOCO和Flickr30k两个数据集的实验结果表明,所提的VRADC方法能够大幅度地提升图像描述生成的质量,实现了最先进的性能.
