Trajectory Obfuscation via Generative State-Space Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for obfuscating user location trajectories fail to generate plausible trajectories that preserve privacy while maintaining utility for location-based applications, often revealing obfuscation due to lack of spatiotemporal correlation and consistency with valid routes and user behavior.
Innovation Solution
A generative state-space model is trained using user location data to create a plausible obfuscated trajectory that respects spatiotemporal correlations, conforms to valid street routes, and preserves user behavior patterns, using a path generator to merge actual and fake locations with minimal distortion, ensuring privacy and utility.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If existing obfuscation methods are used to hide sensitive locations, then privacy is protected, but the trajectory becomes implausible and loses utility for location-based applications
Solution Approach 1:
The patent introduces a generative model as an intermediary that transforms actual trajectories into obfuscated trajectories. This mediator learns the underlying patterns of user behavior and generates synthetic trajectories that preserve these patterns while removing sensitive information, thus maintaining both privacy and utility simultaneously
Solution Approach 2:
The patent changes the parameter representation of trajectories by learning a latent space representation through neural networks. By transforming trajectories into this learned parameter space and then reconstructing them, the system can control the degree of obfuscation while preserving essential characteristics needed for application utility
2Object-affected harmful factors
If simple location obfuscation is applied, then sensitive places are hidden, but spatiotemporal correlation and consistency with valid routes are lost
Solution Approach 1:
The patent employs feedback mechanisms where the generative model is trained on actual trajectories and continuously refines its output to match ground truth patterns. The loss function provides feedback on how well the obfuscated trajectories preserve spatiotemporal correlations and route validity, guiding the model to maintain precision while obfuscating sensitive information
3Productivity
If detailed user location data is preserved for application functionality, then trajectory utility is maintained, but privacy of sensitive places is compromised
Solution Approach 1:
The patent segments the trajectory data into sensitive and non-sensitive components by learning from training data which locations are sensitive. The generative model then selectively obfuscates only the sensitive segments while preserving non-sensitive portions, maintaining application functionality where possible while protecting privacy where needed
Data Source
AI summary
Aspects of the invention include receiving, using a processing system, an actual user location trajectory that includes a plurality of geographic locations of places visited by a user. It is determined that at least one of the plurality of places visited by the user has been identified as a sensitive place. An obfuscated user location trajectory is created that preserves the privacy of the sensitive places that is consistent with the actual user location trajectory that conforms to a valid street route on a map, preserves spatiotemporal correlation between geographic locations, and is consistent with geographic locations visited by the user in the past. Contents of the obfuscated user location trajectory are output to an application in place of contents of the actual user location trajectory.