Publications

Conference Papers

GPS-Enhanced Tourist Mobility Modeling with Seasonal Spatial Priors and LLM-Based Activity Chain Generation

Yifan Liu*, Yanling Sang*, Xishun Liao, Morgan Sun, Bo Yang, Zhiyuan Zhang, Chris Stanford, Haoxuan Ma, Jiaqi Ma

A four-stage framework that generates individual tourist itineraries from month-conditioned GPS spatial priors and LLM-based activity chains, producing synthetic schedules for Tokyo that align with survey and staypoint visitation patterns.

Uncertainty-Aware Trip Purpose Inference from GPS Trajectories via POI Semantic Zones and Pareto Calibration

Bo Yang*, Haoxuan Ma*, Yifan Liu, Zhiyuan Zhang, Chris Stanford, Morgan Sun, Jiaqi Ma

A weakly supervised framework that infers trip purposes from 81M+ GPS staypoints in Los Angeles without labels, combining POI semantic zones with multi-phase Pareto calibration to cut start-time JSD by 48% over a comparable baseline.

Private-RAG: Answering Multiple Queries with LLMs while Keeping Your Data Private

Ruihan Wu*, Erchi Wang*, Zhiyuan Zhang, Yu-Xiang Wang

Two differentially private RAG algorithms (MuRAG and MuRAG-Ada) for the multi-query setting, where privacy loss depends on how often each document is retrieved rather than the total number of queries, scaling to hundreds of queries within a practical budget (ε ≈ 10).

Pre-prints

From One Query to Many: A Critical Survey of Privacy Accounting in RAG

Zhiyuan Zhang

A critical survey of how privacy accounting evolves from single-query to multi-query DP-RAG, proposing that an ε is only meaningful together with its protected unit, accounting scope, and release boundary.

Privacy-Preserving Retrieval-Augmented Generation with Differential Privacy

Tatsuki Koga*, Ruihan Wu*, Zhiyuan Zhang, Kamalika Chaudhuri

A differentially private RAG algorithm that spends privacy budget only on the tokens that need sensitive information and uses the non-private LLM for the rest, outperforming the non-RAG baseline at ε ≈ 10 across models and datasets.