A neural database for differentially private spatial range queries

Mobile apps and location-based services generate large amounts of location data. Location density information from such datasets benefits research on traffic optimization, context-aware notifications public health (e.g., disease spread). To preserve individual privacy, one must sanitize data, which is commonly done using differential privacy (DP). Existing methods partition the data domain into bins, add noise to each bin publish a noisy histogram However, simplistic modelling choices fall short accurately capturing useful in spatial yield poor accuracy. We propose machine-learning based approach for answering range count queries with DP guarantees. focus countering sources error that plague existing approaches (i.e., uniformity error) through learning, we design neural database system models features are preserved, even when DP-compliant added. also devise framework effective parameter tuning top helps set important parameters without expending scarce budget. Extensive experimental results real heterogeneous characteristics show our proposed significantly outperforms state art.