通信学报2025，Vol.46Issue(12)：138-156,19.DOI:10.11959/j.issn.1000-436x.2025252
基于CNN-BiLSTM-Attention融合模型的差分隐私轨迹重构攻击
Trajectory reconstruction attacks on differential privacy based on a CNN-BiLSTM-Attention hybrid model
摘要
Abstract
To address the poor attack performance of existing reconstruction attack methods against differential privacy trajectory protection mechanisms caused by deficiencies in local feature extraction,spatial information extraction,and global dependency modeling,a trajectory reconstruction attack method based on a CNN-BiLSTM-Attention fusion model was proposed.A convolutional neural network(CNN)was introduced to capture spatial dependencies and local patterns in trajectory data.Long term temporal dependencies in trajectory sequences were modeled using bidirectional long short-term memory(BiLSTM)network,thereby strengthening representation capability along the temporal dimen-sion.An attention mechanism was employed to adaptively assign different weights to each time step,capturing global in-formation and long span dependencies within trajectories.Experimental results show that,compared with baseline meth-ods,the average percentage of Euclidean distance reduction of the proposed method is increased by 5.03%,the average improvement in the percentage reduction of Hausdorff distance is 5.02%,and the Jaccard index of the trajectory convex hull increases by an average factor of 2.4,enabling effective trajectory reconstruction attacks.关键词
差分隐私/轨迹重构攻击/卷积神经网络/双向长短期记忆网络/注意力机制Key words
differential privacy/trajectory reconstruction attack/convolutional neural network/bidirectional long short-term memory network/attention mechanism分类
信息技术与安全科学引用本文复制引用
谢丽霞,赵尔康,杨宏宇,刘哲理,赵永新..基于CNN-BiLSTM-Attention融合模型的差分隐私轨迹重构攻击[J].通信学报,2025,46(12):138-156,19.基金项目
国家自然科学基金民航联合研究基金重点项目(No.U2433205)Civil Aviation Joint Research Fund Project of the National Natural Science Foundation of China(No.U2433205) （No.U2433205）
