A Geological Borehole Data Protection Based on Graph Neural Networks
-
摘要: 随着深度学习技术的日益成熟，攻击者可以对公开的地质钻孔数据通过分类、预测等方法获取潜在的敏感信息，从而造成重要地质数据的泄露. 针对上述问题，提出了一种基于图对抗攻击的地质钻孔数据保护模型Gcntack. 一方面，基于地质数据拓扑图的度特征，产生满足同一幂律分布的攻击作为微小节点扰动，确保对抗性攻击不易被发现，同时改变了目标节点的分类结果. 另一方面，引入注意力机制，使用基于可解释性的图注意力网络模型分析影响对抗攻击结果的关键节点特性，验证Gcntack模型中选取对抗性节点的合理性. 最后，通过在基准数据集和地质钻孔数据集进行的综合实验和分析，证实了提出的地质钻孔数据保护方案能够基于较少的图结构或节点特征的对抗扰动，达到保护重要地质钻孔数据的目的.Abstract: With the development of deep learning technology, attackers can obtain potentially sensitive information from public geological data through classification, prediction, and other methods, which could lead to the leakage of important geological data. To solve the above problems, we propose a geological drilling data protection model based on graph adversarial attack Gcntack.Based on the degree properties of geological data topology, we first generate attacks that satisfy the same power-law distribution as tiny node disturbance. It can ensure that the adversarial attacks are not easy to be found, and while can change the classification result of the target node. Secondly, we introduce an attention mechanism. Using a graph attention network model based on interpretability, we analyze the properties of key nodes that directly affect the results of the adversarial attacks, so as to verify the rationality of the selecting adversarial nodes in the Gcntack model. Finally, a comprehensive evaluation, based on the benchmark dataset and geological drilling dataset, is presented to show this proposed scheme can reduce the prediction accuracy of attackers and achieve the purpose of protecting important geological drilling data.
-
图 1 基于图对抗攻击和可解释性的地质钻孔数据保护方案
Fig. 1. Geological borehole data protection scheme based on graph adversarial attack and interpretability
图 3 按照主要矿种-工作区构建的钻孔散点图
Fig. 3. Borehole scatter diagram constructed according to the main ore-working area
图 4 按照终孔深度-钻孔类型构建的钻孔散点图
Fig. 4. Borehole Scatter Diagram constructed according to final hole depth-type of borehole
图 8 7个钻孔类型中注意力权重高的节点与对抗性节点对应的命中率
Fig. 8. The hit ratio of the nodes with high attention weight and the adversarial nodes in the seven drilling types
图 9 排名前10的地质钻孔注意力权重统计结果
Fig. 9. Statistical results of the top 10 geological borehole attention weights
表 1 数据集统计信息
Table 1. Data set statistics
名称 节点数量 边缘数量 特征 类别 Cora 2 708 5 429 1 433 7 Citeseer 3 327 4 732 3 703 6 Drilling-1 670 35 836 106 7 Drilling-2 670 95 472 106 7 
下载: 导出CSV
表 2 数据集上节点分类的准确性（百分比）
Table 2. Accuracy of node classification on dataset (percentage)
Cora Citeseer Drilling-1 Drilling-2 GCN 0.815 0.701 0.709 0.803 GAT 0.845 0.725 0.728 0.826 Gcntack 0.471 0.423 0.410 0.432
下载: 导出CSV
-
Ali-Ozkan, O., Ouda, A., 2019. Key-Based Reversible Data Masking for Business Intelligence Healthcare Analytics Platforms. 2019 International Symposium on Networks, Computers and Communications(ISNCC), 2019: 1-6. https://doi.org/10.1109/ISNCC.2019.8909125 .Bojchevski, A., Günnemann, S., 2019. Adversarial Attacks on Node Embeddings via Graph Poisoning. ICML, 97: 695-704. Borgs, C., Chayes, J., Cohn, H., et al., 2019. An Theory of Sparse Graph Convergence I: Limits, Sparse Random Graph Models, and Power Law Distributions. Transactions of the American Mathematical Society, 372(5): 3019-3062. https://doi.org/10.1090/tran/7543 Cai, H. Y., Zheng, V. W., Chang, K. C. C., 2018. A Comprehensive Survey of Graph Embedding: Problems, Techniques, and Applications. IEEE Transactions on Knowledge and Data Engineering, 30(9): 1616-1637. https://doi.org/10.1109/tkde.2018.2807452 Chen, J. Y., Lin, X., Shi, Z. Q., et al., 2020. Link Prediction Adversarial Attack Via Iterative Gradient Attack. IEEE Transactions on Computational Social Systems, 7(4): 1081-1094. https://doi.org/10.1109/tcss.2020.3004059 Cuzzocrea, A., Shahriar, H., 2017. Data Masking Techniques for NoSQL Database Security: A Systematic Review. 2017 IEEE International Conference on Big Data, 2017: 4467-4473. https://doi.org/10.1109/BigData.2017.8258486 Dai, H.J., Li, H., Tian, T., et al., 2018. Adversarial Attack on Graph Structured Data. ICML, 80: 1123-1132. Eikmeier, N., Gleich, D.F., 2017. Revisiting Power-Law Distributions in Spectra of Real-World Networks. Proceedings of the 23rd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. 2017: 817-826. Gagula, A.C., Santillan, J.R., 2020. Integrating Geographic Information System, Remote Sensing Data, Field Surveys, and Hydraulic Simulations in Irrigation System Evaluation. Proceedings of the 2020 IEEE REGION 10 CONFERENCE (TENCON), Osaka, 2020: 626-630. Hui, Y.U., Wei, Z., Xinnian, M.A., 2017. A Reversible Decryption Model for Vector and Raster Integration Based on Trigonometric Function. Bulletin of Surveying and Mapping, (10): 89-94. Jiang, D.H., Zhou, W., 2018. Decryption Model for Vector Geographic Data Based on Chebyshev Polynomials. Journal of Geomatics Science and Technology. 35(3): 321-325(in Chinese with English abstract). Kipf, T.N., Welling, M., 2017. Semi-Supervised Classification with Graph Convolutional Networks. 5th International Conference on Learning Representations, 1-14. Li, A. B., Zhu, A. X., 2019. Copyright Authentication of Digital Vector Maps Based on Spatial Autocorrelation Indices. Earth Science Informatics, 12(4): 629-639. https://doi.org/10.1007/s12145-019-00386-z Li, Y. F., Jin, R., Luo, Y., 2019. Classifying Relations in Clinical Narratives Using Segment Graph Convolutional and Recurrent Neural Networks (Seg-GCRNs). Journal of the American Medical Informatics Association, 26(3): 262-268. https://doi.org/10.1093/jamia/ocy157 Li, H., Zhu, H.H., Hua, W.H., et al., 2020. Key Technologies and Methods for Vector Geographic Data Security Protection. Earth Science, 45(12): 4574-4588 (in Chinese with English abstract). Liu, H., Zhao, B., Guo, J.B., et al., 2021. Survey on Adversarial Attacks Towards Deep Learning. Journal of Cryptologic Research, 8(2): 202-214(in Chinese with English abstract). Ma, J.Q., Chang, B., Zhang, X., et al., 2020. CopulaGNN: Towards Integrating Representational and Correlational Roles of Graphs in Graph Neural Networks. International Conference on Learning Representations, 2020: 1-13. Marti, R., Li, Z. C., Catry, T., et al., 2020. A Mapping Review on Urban Landscape Factors of Dengue Retrieved from Earth Observation Data, GIS Techniques, and Survey Questionnaires. Remote Sensing, 12(6): 932. https://doi.org/10.3390/rs12060932 Peng, Y. W., Lan, H., Yue, M. L., et al., 2018. Multipurpose Watermarking for Vector Map Protection and Authentication. Multimedia Tools and Applications, 77(6): 7239-7259. https://doi.org/10.1007/s11042-017-4631-z Pham, G.N., Ngo, S.T., Bui, A.N., et al., 2019. Vector Map Random Encryption Algorithm Based on Multi-Scale Simplification and Gaussian Distribution. Applied Sciences, 9(22): 4889. https://doi.org/10.3390/app 9224889 doi: 10.3390/app9224889 Qiu, Y. G., Duan, H. T., Sun, J. Y., et al., 2019. Rich-Information Reversible Watermarking Scheme of Vector Maps. Multimedia Tools and Applications, 78(17): 24955-24977. https://doi.org/10.1007/s11042-019-7681-6 Qiu, Y. G., Gu, H. H., Sun, J. Y., 2017. High-Payload Reversible Watermarking Scheme of Vector Maps. Multimedia Tools and Applications, 77(5): 6385-6403. https://doi.org/10.1007/s11042-017-4546-8 Ren, N., Zhu, C. Q., Tong, D. Y., et al., 2020. Commutative Encryption and Watermarking Algorithm Based on Feature Invariants for Secure Vector Map. IEEE Access, 8: 221481-221493. https://doi.org/10.1109/access.2020.3043450 Sun, Y., Wang, S., Tang, X., et al., 2020. Adversarial Attacks on Graph Neural Networks via Node Injections: A Hierarchical Reinforcement Learning Approach. WWW '20: The Web Conference 2020, 2020: 673-683. Van, B.N., Lee, S.H., Kwon, K.R., 2017. Selective Encryption Algorithm Using Hybrid Transform for GIS Vector Map. Journal of Information Processing Systems, 13(1): 68-82. https://doi.org/10.3745/jips.03.0059 Vybornova, Y., Vladislav, S., 2019. Method for Vector Map Protection Based on Using of a Watermark Image as a Secondary Carrier. Proceedings of the ICETE (2). Prague, Czech Republic, 2019: 284-293. Wang, X.D., Liu, Z., Wang, N.N., et al., 2020. Relational Metric Learning with Dual Graph Attention Networks for Social Recommendation. PAKDD (1) 2020: 104-117. Xia, D., Ge, Y.F., Tang, H.M., et al., 2020. Segmentation of Region of Interest and Identification of Rock Discontinuities in Digital Borehole Images. Earth Science, 45(11): 4207-4217(in Chinese with English abstract). Zhai, M.G., Yang, S.F., Chen, N.H., et al., 2018. Big Data Epoch: Challenges and Opportunities for Geology. Bulletin of Chinese Academy of Sciences, 33(8): 825-831(in Chinese with English abstract). Zhu, D.Y., Zhang, Z.W., Cui, P., et al., 2019. Robust Graph Convolutional Networks Against Adversarial Attacks. Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining. 2019: 1399-1407. Zügner, D., Akbarnejad, A., Günnemann, S., 2018. Adversarial Attacks on Neural Networks for Graph Data. Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, 2018: 2847-2856. 江栋华, 周卫, 2018. 一种基于Chebyshev多项式的矢量数据几何精度脱密模型. 测绘科学技术学报, 35(3): 321-325. https://www.cnki.com.cn/Article/CJFDTOTAL-JFJC201803019.htm 李虎, 朱恒华, 花卫华, 等, 2020. 矢量地理数据安全保护关键技术和方法. 地球科学, 45(12): 4574-4588. doi: 10.3799/dqkx.2020.299 刘会, 赵波, 郭嘉宝, 等, 2021. 针对深度学习的对抗攻击综述. 密码学报, 8(2): 202-214. https://www.cnki.com.cn/Article/CJFDTOTAL-MMXB202102002.htm 夏丁, 葛云峰, 唐辉明, 等, 2020. 数字钻孔图像兴趣区域分割与岩体结构面特征识别. 地球科学, 45(11): 4207-4217. doi: 10.3799/dqkx.2020.003 翟明国, 杨树锋, 陈宁华, 等, 2018. 大数据时代: 地质学的挑战与机遇. 中国科学院院刊, 33(8): 825-831. https://www.cnki.com.cn/Article/CJFDTOTAL-KYYX201808012.htm -
点击查看大图
图(9) / 表(2)
计量
- 文章访问数: 305
- HTML全文浏览量: 486
- PDF下载量: 36
- 被引次数: 0
