ID	原文	译文
2043	实验结果表明了所提算法在性能和训练时间上取得了良好的效果。	Experimental results show that the proposed algorithm achieves good results in terms of performance and training time.
2044	现有行为识别方法在未能持续覆盖造成视频监控盲区所引起行为数据缺失的情况，难以有效实施特征分析、行为分类补全，无法准确识别出智能体完整的行为动作序列。	Currently, action recognition methods can hardly carry out feature analysis, behavior classification, and ac-tion completion, and are incapable of accurately identifying the complete behavioral action sequence of intelligent agent forthe discontinuous and incomplete motion capture, behavioral data missing or even broken in the time dimension, which are resulted from sensor device not being continuous coverage caused by the monitoring blind area.
2045	为此，本文提出一种基于深度学习和智能规划的行为识别方法。	In this regard, we put for-ward a method of action recognition based on deep learning and artificial intelligence planning.
2046	首先，利用深度残差网络对图像进行分类训练，然后使用递归神经网络对图像特征进行提取深度信息以增强分类效果;	Firstly, a deep learning net-work is constructed, by which the image is classified and trained using DRN(Deep Residual Network). After that, the ex-traction depth information of image frame feature for recurrent neural network is trained to enhance the classification effect.
2047	其次，运用智能规划的 STRIPS(Stanford Research Institute Problem Solver)模型，将深度学习提取的图像特征命题信息转化为规划领域的模型描述文档，并使用前向状态空间搜索规划器推导出完整的行为动作序列。	Secondly, the STRIPS(Stanford Research Institute Problem Solver)planning model is used to extract the image feature ofdeep learning, transforming into the description document for domain model, which facilitates deriving the optimal planning solution by means of forward state-space search planner.
2048	在HMDB51 等行为识别公共数据集中，本方法与生成式对抗网络、深度卷积逆向图网络、深度信念网络、支持向量机等同类先进方法相比展现出更好的性能。	In the experiment, we exhibit that our method outperforms baselines in the public datasets, e. g. , DCIGN(Deep Convolutional Inverse Graphics Networks), GAN(Generative Adversarial Net-works), DBN(Deep Belief Networks), and SVM(Support Vector Machine).
2049	针对当前聚类方法(例如经典的 GN 算法)计算复杂度过高、难以适用于大规模图的聚类问题，本文首先对大规模图的采样算法展开研究，提出了能够有效保持原始图聚类结构的图采样算法(Clustering-structure Repre-sentative Sampling，CRS)，它能在采样图中产生高质量的聚类代表点，并根据相应的扩张准则进行采样扩张。	Since computational complexities of the existing methods such as classic GN algorithm are too costly tocluster large-scale graphs, this paper studies sampling algorithms of large-scale graphs, and proposes a clustering-structure representative sampling (CRS)which can effectively maintain the clustering structure of original graphs.
2050	此采样算法能够很好地保持原始图的内在聚类结构。	It can produce highquality clustering-representative nodes in samples and expand according to the corresponding expansion criteria.
2051	其次，提出快速的整体样本聚类推断(Population Clustering Inference，PCI)算法，它利用采样子图的聚类标签对整体图的聚类结构进行推断。	Then, wepropose a fast population clustering inference (PCI)method on the original graphs and deduce clustering assignments of the population using the clustering labels of the sampled subgraph.
2052	实验结果表明本文算法对大规模图数据具有较高的聚类质量和处理效率，能够很好地完成大规模图的聚类任务。	Experiment results show that in comparison with state-of-the-art methods, the proposed algorithm achieves better efficiency as well as clustering accuracy on large-scale graphs.