[1]孔秀平,陆 林.隐私保护下的车辆轨迹联邦嵌入学习与聚类[J].南京师范大学学报(工程技术版),2022,(02):080-86.[doi:10.3969/j.issn.1672-1292.2022.02.012]
 Kong Xiuping,Lu Lin.Privacy-preserved Vehicular Trajectory Embedding Federated Learning and Clustering[J].Journal of Nanjing Normal University(Engineering and Technology),2022,(02):080-86.[doi:10.3969/j.issn.1672-1292.2022.02.012]
点击复制

隐私保护下的车辆轨迹联邦嵌入学习与聚类
分享到:

南京师范大学学报(工程技术版)[ISSN:1006-6977/CN:61-1281/TN]

卷:
期数:
2022年02期
页码:
080-86
栏目:
计算机科学与技术
出版日期:
2022-06-30

文章信息/Info

Title:
Privacy-preserved Vehicular Trajectory Embedding Federated Learning and Clustering
文章编号:
1672-1292(2022)02-0080-07
作者:
孔秀平1陆 林2
(1.扬州工业职业技术学院信息中心,江苏 扬州 225127)(2.中电云数智科技有限公司,湖北 武汉 430056)
Author(s):
Kong Xiuping1Lu Lin2
(1.Department of Information Center,Yangzhou Polytechnic Institute,Yangzhou 225127,China)(2.China Electronic Cloud Digital Intelligence Technology Co.,Ltd.,Wuhan 430056,China)
关键词:
序列自编码联邦学习差分隐私轨迹聚类
Keywords:
sequential autoencoderfederated learningdifferential privacytrajectory clustering
分类号:
TP311.1
DOI:
10.3969/j.issn.1672-1292.2022.02.012
文献标志码:
A
摘要:
智能网联汽车的高维轨迹数据被广泛用于从车辆的行驶轨迹中发现不同运动模式,从而降低交通风险、提高通行效率. 然而,数据利用过程中的隐私问题日益受到关注,如何在隐私保护的前提下进行算法的研究和应用是当前面临的一大挑战. 针对车辆轨迹数据分散在不同持有方且出于隐私保护无法共享数据的背景,利用差分隐私联邦学习框架来构建序列自编码网络提取轨迹序列的低维表示,并进一步利用轨迹的低维空间向量来发现不同时段下车辆的频繁路线. 提出的框架既通过本地训练避免了用户隐私数据的分享,又能通过高斯差分隐私机制防止模型信息的泄露. 该框架在真实的轨迹数据集上进行了验证,利用LSTM自编码作为嵌入学习网络,与非联邦、非差分加密的模型进行了对比分析,最后对三种得到的轨迹嵌入通过聚类分析发现该框架下学习的模型在充分尊重了隐私保护的前提下,仍然能够找出有效的频繁轨迹.
Abstract:
High dimensional trajectory data of intelligent networked vehicles are widely used to find different motion patterns from vehicle trajectories,so as to reduce traffic risk and improve traffic efficiency. However,more and more attention has been paid to the privacy problem in the process of data utilization. How to research and apply the algorithm under the premise of privacy protection is a big challenge. In view of the background that vehicle trajectory data are scattered among different owners and cannot be shared due to privacy protection,this paper uses differential privacy federated learning framework to construct a sequence autoencoding network to extracting low dimensional representation of trajectory sequence,and such latent representation is further used to find frequent vehicle routes in different periods. The proposed framework not only avoids the sharing of user privacy data through local training,but also prevents the disclosure of model information through Gaussian differential privacy mechanism. The framework is validated on real trajectory data sets,using LSTM autoencoder as embedding learning network,and compared with non-federated and non-differential encryption models. Finally,through clustering analysis,it is found that the learning model under the framework can still find effective frequent trajectories under the premise of fully respecting privacy protection.

参考文献/References:

[1] ATEV S,MILLER G,PAPANIKOLOPOULOS N P. Clustering of vehicle trajectories[J]. IEEE Transactions on Intelligent Transportation Systems,2010,11(3):647-657.
[2]BIAN J,TIAN D Y,TANG Y Y,et al. A survey on trajectory clustering analysis[J]. arXiv Preprint,2018:1-40.
[3]柳盛,吉根林. 空间聚类技术研究综述[J]. 南京师范大学学报(工程技术版),2010,10(2):57-62.
[4]陈铭,吉根林. 一种基于相似维的高维子空间聚类算法[J]. 南京师大学报(自然科学版),2010,33(4):119-122.
[5]BRELL T,BIERMANN H,PHILIPSEN R,et al. Conditional privacy:users’ perception of data privacy in autonomous driving[C]//Proceedings of the 5th International Conference on Vehicle Technology and Intelligent Transport Systems. Anchorage,USA:IEEE,2019:352-359.
[6]姚瑶,吉根林. 面向垂直划分数据库的隐私保护分布式聚类算法[J]. 南京师范大学学报(工程技术版),2008,8(4):099-102.
[7]KYUNGHY N C,BART V M,CAGLAR G,et al. Learning phrase representations using RNN encoder-decoder for statistical machine translation[J]. arXiv Preprint arXiv:1406.1078,2014.
[8]YAO D,ZHANG C,ZHU Z H,et al. Trajectory clustering via deep representation learning[C]//International Joint Conference on Neural Networks. Anchorage,USA:IEEE,2017.
[9]YUE M X,LI Y G,YANG H Z,et al. DETECT:deep trajectory clustering for mobility-behavior analysis[C]//2019 IEEE Internation Conference on Big Data,Los angeles,USA:IEEE,2019.
[10]张新峰,闫昆鹏,赵珣. 基于双向LSTM的手写文字识别技术研究[J]. 南京师大学报(自然科学版),2019,42(3):58-64.
[11]杨强,刘洋,陈天健,等. 联邦学习[M]. 北京:电子工业出版社,2020.
[12]YANG Q,LIU Y,CHEN T J,et al. Federated machine learning:concept and applications[J]. ACM Transactions on Intelligent System Technology,2019 10(2):1-19.
[13]DWORK C,ROTH A. The algorithmic foundations of differential privacy[J]. Foundations and Trends in Theoretical Computer Science,2014,9(3/4):211-407.
[14]ABADI M,CHU A,GOODFELLOW I,et al. Deep learning with differential privacy[C]//Proceedings of the 2016 ACM SIGSAC Conference on Computer and Communications Security. New York,USA:Association for Computing Machinery. 2016:308-318.
[15]BU Z Q,DONG J S,LONG Q,et al. Deep learning with Gaussian differential privacy[J]. arXiv Preprint arXiv:1911.11607,2019.
[16]MIRONOV I,TALWAR K,ZHANG L. Rényi differential privacy of the sampled Gaussian mechanism[J]. arXiv Preprint arXiv:1908.10530,2019.

备注/Memo

备注/Memo:
收稿日期:2021-08-31.
基金项目:国家自然科学基金面上项目(61902070).
通讯作者:孔秀平,硕士,工程师,研究方向:计算机网络与隐私保护. E-mail:xiu651015722@qq.com
更新日期/Last Update: 1900-01-01