ID 原文 译文
40826 本文提出一种面向不平衡数据分类的高维超球体过采样(HS-SMOTE)方法。 A high-dimensional Hypersphere-SMOTE(HS-SMOTE) method isproposed for imbalanced data classification.
40827 在少数类样本集上通过随机抽样获得需要平衡的样本数目，在此基础上依次对每一样本通过欧氏距离选取其在少数类分布空间中的对应最近邻点，以两点连线中点为球心在超维空间构建采样超球体， On the minority sample set, the number of samples that need to be balanced is obtained by random sampling, and based on this sampling, for each sample, its corresponding nearest neighbor is selected in turn through the Euclidean distance in the minority distribution space, and the midpoint of the two points is used for the center to construct a sampled hypersphere in the super-dimensional space.
40828 在此区域内通过维度空间距离迭代随机生成所需的少数类新点，在类别样本数据再平衡的基础上增加少数类样本的空间分布多样性。 In this area, randomly generate the required minority new points through the dimensional space distance iteration, thus the spatial distribution diversity of the minority samples is increased on the basis of rebalancing the category sample data.
40829 在15组KEEL不平衡数据集上结合随机森林(RF)分类器开展了大量实验，与6种典型过采样方法相比，所提方法在G-mean以及F1-score指标上均有较好的表现，并通过了2种统计学假设检验方法的有效性验证。 A large number of experiments have been carried out on 15 sets of KEEL imbalanced data sets combining Random Forest(RF)classifiers.Compared with the 6 typical oversampling methods, the method proposed in the article has good performance on G-meanandF1-score indicators, and have passed the validity verification of two statistical hypothesis testing methods.
40830 随着深度学习和神经网络技术的发展，为了充分挖掘卷积神经网络(CNN)计算的并行性，硬件加速器以其高速、低成本、高容错能力等特点得到更加广泛的应用。 In order to fully explore the parallelism of convolutional neural network(CNN) computing, hardware accelerators are more attractive for their characteristics of high speed, low cost and high fault tolerance.
40831 本文提出了一种可以逐层优化CNN网络的新算法，设计了对应的指令集。 A novel algorithm that can optimize the CNN network layer by layeris proposed, and the corresponding instruction set is designed in this paper.
40832 所提出的算法可用于为具有特定计算资源和存储资源的不同网络找到最佳加速方案。 The proposed algorithm can be used to find an optimal acceleration scheme for differ-ent networks with specific computing and storage resources.
40833 在优化过程中，可以将不同类型的数据量化为半精度以减少内存访问。 In the optimization process, different types of data can be quantized to half-precision to reduce memory access.
40834 基于40 nm CMOS工艺和提出的算法，完成了一种指令集控制的神经网络加速器设计。该加速器在200 MHz的工作频率下，峰值性能可达到416 GOP/s。 Based on the 40 nm CMOS process and the proposed algorithm, aprogrammable accelerator for CNN is designed, which can achieve peak performance of 416 GOP/s under 200 MHz working frequency.
40835 在设计的加速器上实现了VGG16网络的推理过程，整个网络的延迟仅为116毫秒。 VGG is implemented on our accelerator as a case study, and the latency of the total network is 116 ms.