ID 原文 译文
25535 TSL 同时解决了传统 TCP Incast 问题和多轮数据传输下由遗留窗口引发的 TCP Incast 问题。 TSL solves the traditional TCP Incast problem and the TCP Incast problem caused by the legacy window in multi-round of data transmission．
25536 实验表明，TSL 在单轮数据传输和多轮数据传输下均能获得 90% 以上的带宽利用率。 Through extensive experiments we show that TSL can achieve more than 90% goodput regardless of single or multiple rounds of data transmission．
25537 在 10Gbps 网络中，其支持的并发连接数与传统 TCP 和 DCTCP 相比分别提升了 5 倍和 1 倍，有效吞吐率分别提升了 18倍和 8. 6 倍; In a 10Gbps network, the number of concurrent connections supported by TSL has increased by 5 times and once respectively compared to the traditional TCP and DCTCP, and the goodput has increased by 18 times and 8. 6 times respectively．
25538 在 1Gbps 网路中，支持的并发连接数较传统 TCP 和 DCTCP 分别提升了 5. 8 倍和 1 倍。 In a 1Gbps network, the number of concurrent connections supported by TSL has increased by 5. 8 times and once respectively compared to the traditional TCP and DCTCP.
25539 基于元胞自动机(CA)的 S 盒密码学性质良好且软硬件实现代价低，被用于 Keccak、SIMON 等密码算法。 Cellular automata (CA) based S-boxes are the type of S-boxes with good cryptography and low cost of hardware as well as software implementation, which are used in Keccak, SIMON, and other cryptographic algorithms.
25540 本文研究了基于 CA 的 S 盒的性质，给出并证明了此类 S 盒的三个重要性质:移位不变性、镜面对称性和互补性; This paper studied the properties of CA-based S-boxes, and the three important properties were given and proved, including shifti nvariance, mirror symmetry and complementarity．
25541 同时研究了基于 CA 的 S 盒的神经网络实现方法，指出相比一般的 S 盒，基于 CA 的 S 盒在进行神经网络实现时可以用更简单的结构、消耗更少的资源来完成，并且给出了一种权重阈值搜索算法可以方便快速地实现基于 CA 的 S 盒的神经网络结构。 Meanwhile, the neural network implementation for CA-based S-boxes was studied, which demonstrated that the CA-based S-boxes could be implemented with simpler structure and less resources than the general one. In addition, a weight threshold search algorithm which could easily and quickly implement the neural network structure of CA-based S-boxes was shown．
25542 现有深度卷积神经网络中感受野尺度单一，无法适应目标的尺度变化和边界形变，故此本文提出了一种提取并融合多尺度特征的目标检测网络。 In the existing deep convolution neural network, the scale of receptive field is single, which could not adapt to the scale change and boundary deformation of the target. Therefore, a target detection network based on multi-scale feature extraction and feature fusion is proposed in this paper．
25543 该网络通过减少池化并在网络底层加入空间加信道压缩激励模块来突出可利用的细节信息，生成高质量的特征图; The proposed network reduces pooling and adds space as well as channel compression excitation module at the bottom of the network to highlight the details and generate high-quality feature map．
25544 此外，在深层网络中加入可变多尺度特征融合模块，该模块具有多种尺度的感受野并可根据物体边界预测采样位置，最后通过融合多尺度特征使网络具有更强的特征表达能力并且对不同尺度实例及其边界信息更具鲁棒性。 Besides, a variable multi-scale feature fusion module is added to the deep network, which has a multi-scale receptive field and can predict the position according to object boundary. Finally, the multi-scale feature fusion is used to enable the network of stronger ability of feature expression and is more robust to different scale and flexible boundary of instances．