Reproduce “Generating synthetic mobility data for a realistic population with RNNs to improve utility and privacy” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Alex Berke, Ronan Doorley, Kent Larson, Esteban Moro
Researchers at MIT Media Lab developed an RNN-based system to generate synthetic mobility traces for realistic populations, addressing limitations in sample size and privacy concerns of real location data. The system produces high-utility synthetic data that closely matches real-world mobility patterns while introducing sufficient individual-level variation to enhance privacy.
Key insights
- RNNs, specifically LSTMs with attention, can effectively capture and reproduce the intricate, routine-based, and sequential patterns of human mobility over extended periods, exceeding simpler statistical models.
- Introducing a controlled amount of randomness during the RNN generation process allows for a balance between maintaining high data utility (statistical fidelity) and enhancing privacy (individual variation).
- Prefixing mobility trajectories with inferred home and work locations provides a robust mechanism to generate synthetic populations of arbitrary size that are representative of real demographic distributions.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.