Abstract:Human mobility data offers valuable insights for many applications such as urban planning and pandemic response, but its use also raises privacy concerns. In this paper, we introduce the Hierarchical and Multi-Resolution Network (HRNet), a novel deep generative model specifically designed to synthesize realistic human mobility data while guaranteeing differential privacy. We first identify the key difficulties inherent in learning human mobility data under differential privacy. In response to these challenges, HRNet integrates three components: a hierarchical location encoding mechanism, multi-task learning across multiple resolutions, and private pre-training. These elements collectively enhance the model's ability under the constraints of differential privacy. Through extensive comparative experiments utilizing a real-world dataset, HRNet demonstrates a marked improvement over existing methods in balancing the utility-privacy trade-off.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is how to improve the practicality and accuracy of generated data while ensuring differential privacy (DP) when generating human mobility data. Specifically, the paper addresses two main bottlenecks encountered by existing methods when dealing with large - scale and complex human mobility data: 1. **Excessive number of model parameters**: As the number of POIs (points of interest) increases, the number of model parameters grows linearly, leading to an increase in empirical risk under differential privacy, thereby affecting the practicality of the generated data. 2. **Excessive number of labels**: As the number of POIs increases, the model needs to distinguish more subtle feature differences, which usually requires a larger model architecture or a longer training period, but this is impractical under DP - SGD constraints. Moreover, increasing the number of training rounds will lead to higher privacy loss. To solve these problems, the paper proposes HRNet (Hierarchical and Multi - Resolution Network), which is a novel deep - generation model specifically designed to synthesize realistic human mobility data with multi - level resolutions while ensuring differential privacy. HRNet addresses the above challenges through the following three key components: 1. **Hierarchical position encoding mechanism**: A transposed convolutional network is used to reduce the number of parameters, reducing the parameter complexity from \(O(n_{\text{POI}})\) to \(O(\log n_{\text{POI}})\), thus alleviating the problem of excessive parameters. 2. **Multi - task learning**: By performing multi - task learning at multiple resolutions, the model can infer complex fine - grained tasks from coarser - grained tasks, effectively alleviating the learning difficulties caused by an excessive number of labels. 3. **Private pre - training**: By using a differentially - privacy - compatible dense transformation matrix for pre - training, a "warm - start" effect is provided, reducing the dependence on a large number of training rounds and further improving the model performance. Through these innovations, HRNet significantly improves the quality and practicality of the generated data while ensuring privacy.

HRNet: Differentially Private Hierarchical and Multi-Resolution Network for Human Mobility Data Synthesization

HRNet: Differentially Private Hierarchical and Multi-Resolution Network for Human Mobility Data Synthesization

GANs Based Density Distribution Privacy-Preservation on Mobility Data

DPHMM: Customizable Data Release with Differential Privacy Via Hidden Markov Model.

Federated Representation Learning with Data Heterogeneity for Human Mobility Prediction

A Multi-Scale Unified Model of Human Mobility in Urban Agglomerations

Modeling Real-Time Human Mobility Based on Mobile Phone and Transportation Data Fusion

Exploring Multi-Scale Spatiotemporal Twitter User Mobility Patterns with A Visual-Analytics Approach

Privacy-Aware Adversarial Network in Human Mobility Prediction

Generating synthetic mobility data for a realistic population with RNNs to improve utility and privacy

Differential Privacy in Aggregated Mobility Networks: Balancing Privacy and Utility

PateGail: A Privacy-Preserving Mobility Trajectory Generator with Imitation Learning

Anonymization and De-Anonymization of Mobility Trajectories: Dissecting the Gaps Between Theory and Practice

Privacy-Aware Human Mobility Prediction via Adversarial Networks

Nationwide synthetic human mobility dataset construction from limited travel surveys and open data

Hierarchical fuzzy neural networks with privacy preservation for heterogeneous big data

Activity-aware Human Mobility Prediction with Hierarchical Graph Attention Recurrent Network

PMF: A Privacy-preserving Human Mobility Prediction Framework via Federated Learning

Restoring Super-High Resolution GPS Mobility Data

Metropolitan Scale and Longitudinal Dataset of Anonymized Human Mobility Trajectories

Learning to Generate Pseudo Personal Mobility