ID 原文 译文
16905 随机梯度下降算法(SGD)随机使用一个样本估计梯度，造成较大的方差，使机器学习模型收敛减慢且训练不稳定。 The Stochastic Gradient Descent (SGD) algorithm randomly picks up a sample to estimate gradients,creating big variance which reduces the convergence speed and makes the training unstable.
16906 该文提出一种基于方差缩减的分布式SGD，命名为DisSAGD。 A Distributed SGDbased on Average variance reduction, called DisSAGD is proposed.
16907 该方法采用历史梯度平均方差缩减来更新机器学习模型中的参数，不需要完全梯度计算或额外存储，而是通过使用异步通信协议来共享跨节点的参数。 The method uses the average variancereduction based on historical gradients to update parameters in the machine learning model, requiring littlegradient calculation and additional storage, but using the asynchronous communication protocol to shareparameters across nodes.
16908 为了解决全局参数分发存在的“更新滞后”问题，该文采用具有加速因子的学习速率和自适应采样策略： In order to solve the “update staleness” problem of global parameter distribution, alearning rate with an acceleration factor and an adaptive sampling strategy are included:
16909 一方面当参数偏离最优值时，增大加速因子，加快收敛速度； on the one hand,when the parameter deviates from the optimal value, the acceleration factor is increased to speed up theconvergence;
16910 另一方面，当一个工作节点比其他工作节点快时，为下一次迭代采样更多样本，使工作节点有更多时间来计算局部梯度。 on the other hand, when one work node is faster than the other ones, more samples are sampledfor the next iteration, so that the node has more time to calculate the local gradient.
16911 实验表明：DisSAGD显著减少了循环迭代的等待时间，加速了算法的收敛，其收敛速度比对照方法更快，在分布式集群中可以获得近似线性的加速。 Experiments show that the DisSAGD reduces significantly the waiting time of loop iterations, accelerates the convergence of the algorithm being faster than that of the controlled methods, and obtains almost linear acceleration in distributed cluster environments.
16912 基于成像场景散射强度稀疏表示的3维雷达成像结果对目标的外形几何细节体现较差，不利于目标识别。 The three-Dimensional (3D) radar imaging methods based on sparse representation by the scattering intensity of imaging sceen has a poor representation of geometric details on the shape of the target, which isn’t conducive to target recognition.
16913 该文首先分析了目标在成像场景内散射强度的结构化特征， Firstly, the structural characteristics of scattering intensity in the imagingscenario are analyzed in this paper.
16914 然后以散射点梯度信息进行了结构化稀疏表示，构建了基于目标散射强度梯度变化的结构化稀疏重构模型， Then, by the structured sparse representation with the gradient informationof scattering points, a structured sparse reconstruction model is constructed.