ID 原文 译文
25505 本文首次提出了一个基于神经网络的多集群分布式差分隐私数据发布方法，能够显著缓解单服务器的数据处理压力．同时，利用神经网络算法进行隐私参数预测明显提高了预测精度和预测效率，并且集群之间不同的隐私参数也保证了方案的灵活性。 Aiming at the significant relief of data processing pressure on single server, we propose a multi-cluster distributed differential privacy data publishing method based on neural network (MCDP), which effectively improves the prediction accuracy and efficiency, and different privacy parameters between clusters guarantee the flexibility of the protocol．
25506 此外，由于中心服务器存储的是经过差分隐私处理后的统计数据，即使中心服务器由于遭受攻击导致存储的数据泄露，也能确保用户数据隐私。 Especially, because the central server stores statistical data after differential privacy processing, it does not collect individual privacy data, even if the central server is attacked, the user data will not be leaked.
25507 实验对比分析表明，我们的方法在隐私处理效率、隐私保护强度、预测精度和预测效率等方面都有明显优势。 Experiments show that MCDP has obvious advantages in privacy processing efficiency, privacy protection intensity, prediction accuracy and prediction efficiency．
25508 图数据的处理面临庞大规模和复杂结构的制约。 The processing of graph data is restricted by large scale and complex structure．
25509 图的概要化，旨在寻找一组简洁的超图或稀疏图，阐明原始图的主要结构信息或变化趋势。 The graph summarization is to find a set of simple hyper-graphs or sparse graphs that clarify the main structural information or change trend of the original graph．
25510 针对属性图提出了基于编辑行为码的概要模型，遵循最小描述长度原理(Mini-mum Description Length，MDL)，将结构的相似性和属性的相似性统一为存储代价，构建编辑行为码。 According to the principle of minimum description length (MDL), a summary model based on edit coding is proposed for attribute graphs, which unifies the similarity of structure and attribute as storage cost to construct editing behavior code．
25511 在此模型基础上提出了 Greedy 算法和 Random 算法，存储属性和结构的编辑信息，生成高质量的超图，并支持原始图的重构。 Based on this model, a greedy algorithm and a random algorithm are proposed, which store the editing information of attribute and structure, generate high-quality hyper-graphs, and support the reconstruction of original graphs．
25512 实验结果表明本文提出的概要模型和算法相比于其他图概要算法，在压缩率和时间代价等指标上具有一定的优越性。 The experimental results show that the proposed model and algorithms in this paper have some advantages over other summarization algorithms in terms of compression rate and time cost．
25513 针对传统稀疏解混方法对丰度的稀疏性表征不充分及空间信息利用率低等问题，本文在分析迭代加权稀疏解混方法的基础上，提出了一种基于光谱加权协同稀疏和全变差正则化的高光谱解混方法。 In this work, we proposed a hyperspectral unmixing method based on the spectrally weighted collaborative sparsity and the total variation, aiming at alleviating the lack of the sparsity of abundance in traditional methods and fully exploiting the spatial information．
25514 该方法一方面在协同稀疏解混的基础上引入光谱加权因子进一步刻画丰度系数的行稀疏性，以促进所有像元之间的联合稀疏性; On the one hand, the spectral factors are utilized to estimate the weights in order to enforce the sparsity of nonzero rows, thus improving the collaborative sparsity among all the pixels．