diffGHOST: Diffusion based Generative Hedged Oblivious Synthetic Trajectories

diffGHOST is a new conditional diffusion model that synthesizes mobility trajectories while preserving privacy through latent space segmentation. The approach addresses a critical gap in existing generative models that lack formal privacy guarantees despite handling sensitive personal movement data.

The proliferation of location data and trajectory information creates a fundamental tension between utility and privacy. Traditional generative models assume that synthetic data generation inherently provides privacy protection, a flawed premise that leaves sensitive information vulnerable to inference attacks and memorization of training samples. diffGHOST tackles this problem by introducing a diffusion-based architecture with latent space segmentation, enabling the model to identify and mitigate memorization of critical samples—those most likely to compromise individual privacy.

Trajectory data applications span urban planning, transportation optimization, epidemiological modeling, and location analytics. The inability to safely share this data has constrained research and commercial applications in these domains. Previous approaches either sacrificed utility for privacy or provided neither guarantee. The conditional diffusion framework with learned latent space segmentation represents a meaningful advancement in differentially private synthetic data generation, addressing a known limitation in the field.

The technical contribution matters for both researchers and enterprises handling sensitive mobility datasets. Organizations in transportation, telecommunications, and urban development can benefit from generating privacy-preserving synthetic trajectories that retain statistical properties of original data. This enables collaboration and research without exposing individuals' movement patterns.

The impact extends to regulatory compliance contexts where data protection regulations increasingly scrutinize location tracking. As privacy expectations tighten globally, methods that provide formal guarantees become competitive advantages. Future work will likely focus on computational efficiency, scalability to large trajectory datasets, and formal privacy verification across diverse geographic contexts.

• →diffGHOST uses conditional diffusion models with latent space segmentation to generate privacy-preserving synthetic trajectories
• →The model explicitly addresses memorization of critical samples, providing privacy guarantees traditional generative approaches lack
• →Privacy-preserving trajectory synthesis enables safe data sharing for urban planning, transportation, and research applications
• →The approach bridges the gap between utility and privacy by combining formal privacy mechanisms with synthetic data quality
• →Applicable to regulatory compliance contexts where location data protection requirements are increasingly stringent