A multicellular DCX (dc-dc transformer) using unregulated cell converters has been proposed for the environmentally friendly data centers. The high speed cell converter with the switching frequency over MHz behaves ...A multicellular DCX (dc-dc transformer) using unregulated cell converters has been proposed for the environmentally friendly data centers. The high speed cell converter with the switching frequency over MHz behaves as an ideal transformer, and this behavior solves the voltage imbalance issue in the multicellular converter topology. The analysis of the unregulated cell converter is conducted by using the state space averaging method, and the operation condition for the ideal transformer is specified. The behavior of the multicellular DCX using the high speed cell converters has been also analyzed, and the voltage imbalance issue among cell converters is discussed quantitatively. A prototype of a 19.2 kW 384 V-384 V multicellular DCX using sixty-four unregulated cell converters is fabricated and the validity of the analyses is verified.展开更多
针对主流Transformer网络仅对输入像素块做自注意力计算而忽略了不同像素块间的信息交互,以及输入尺度单一导致局部特征细节模糊的问题,本文提出一种基于Transformer并用于处理视觉任务的主干网络ConvFormer. ConvFormer通过所设计的多...针对主流Transformer网络仅对输入像素块做自注意力计算而忽略了不同像素块间的信息交互,以及输入尺度单一导致局部特征细节模糊的问题,本文提出一种基于Transformer并用于处理视觉任务的主干网络ConvFormer. ConvFormer通过所设计的多尺度混洗自注意力模块(Channel-Shuffle and Multi-Scale attention,CSMS)和动态相对位置编码模块(Dynamic Relative Position Coding,DRPC)来聚合多尺度像素块间的语义信息,并在前馈网络中引入深度卷积提高网络的局部建模能力.在公开数据集ImageNet-1K,COCO 2017和ADE20K上分别进行图像分类、目标检测和语义分割实验,ConvFormer-Tiny与不同视觉任务中同量级最优网络RetNetY-4G,Swin-Tiny和ResNet50对比,精度分别提高0.3%,1.4%和0.5%.展开更多
文摘A multicellular DCX (dc-dc transformer) using unregulated cell converters has been proposed for the environmentally friendly data centers. The high speed cell converter with the switching frequency over MHz behaves as an ideal transformer, and this behavior solves the voltage imbalance issue in the multicellular converter topology. The analysis of the unregulated cell converter is conducted by using the state space averaging method, and the operation condition for the ideal transformer is specified. The behavior of the multicellular DCX using the high speed cell converters has been also analyzed, and the voltage imbalance issue among cell converters is discussed quantitatively. A prototype of a 19.2 kW 384 V-384 V multicellular DCX using sixty-four unregulated cell converters is fabricated and the validity of the analyses is verified.
文摘针对主流Transformer网络仅对输入像素块做自注意力计算而忽略了不同像素块间的信息交互,以及输入尺度单一导致局部特征细节模糊的问题,本文提出一种基于Transformer并用于处理视觉任务的主干网络ConvFormer. ConvFormer通过所设计的多尺度混洗自注意力模块(Channel-Shuffle and Multi-Scale attention,CSMS)和动态相对位置编码模块(Dynamic Relative Position Coding,DRPC)来聚合多尺度像素块间的语义信息,并在前馈网络中引入深度卷积提高网络的局部建模能力.在公开数据集ImageNet-1K,COCO 2017和ADE20K上分别进行图像分类、目标检测和语义分割实验,ConvFormer-Tiny与不同视觉任务中同量级最优网络RetNetY-4G,Swin-Tiny和ResNet50对比,精度分别提高0.3%,1.4%和0.5%.