580 California St., Suite 400
San Francisco, CA, 94104
Key research themes
1. How can machine learning and similarity-based models improve short-term location predictability in mobile networks?
This research theme focuses on the application of supervised learning and classification algorithms to predict the immediate or next location of mobile users based on their recent movement histories and spatial context. It addresses the challenges of capturing user mobility patterns with scalability and accuracy, exploiting temporal, spatial, and velocity features, and evaluating predictive performance against baseline mobility models. The work is relevant for enhancing wireless network protocols, location-based services (LBS), and mobile context-aware applications that benefit from proactive location estimations.
Key finding: Proposes a supervised learning-based trajectory classifier and a location predictor model leveraging short-term spatiotemporal user mobility history and neighboring cell information. Their approach achieves increased... Read more
Key finding: Introduces an online trajectory classification system incorporating a novel weighted spatial and velocity context distance metric to differentiate between spatially similar but velocity-different movement patterns. The model... Read more
Key finding: Develops a spatial context model for location prediction treated as a supervised classification task, comparing several machine learning algorithms including decision trees, k-nearest neighbors, and ensemble methods. Their... Read more
Key finding: Extends the Mobility Markov Chain (MMC) model to include n previously visited locations, movement factors, and temporal information (time of day and day) for scalable next-location prediction. Incorporating temporal and... Read more
2. What strategies optimize energy-efficient location sampling tailored to individual mobility patterns?
This theme explores methods for adaptive and energy-aware tracking of mobile users by adjusting the frequency of location sampling according to the regularity and predictability of individual movements. It investigates how spatiotemporal regularity in human mobility enables a trade-off between sampling rate and localization accuracy, aiming to minimize energy consumption without sacrificing location quality. These strategies have implications for mobile computing, reducing battery drain, and preserving privacy by limiting unnecessary data collection.
Key finding: Unveils a novel linear scaling law relating localization error to sampling interval duration and leverages it to adaptively adjust sampling frequency per user mobility profile. The DuctiLoc system saves 60% to 98% energy... Read more
Key finding: Presents two complementary algorithms: the Critical Point (CP) algorithm to reduce the number of GPS fixes transmitted to servers, and a location-aware state machine to dynamically adjust position update intervals. Together... Read more
3. How can social and contextual semantics be leveraged and protected in location predictability and privacy?
This theme investigates the dual role of social network ties and contextual location semantics in influencing location predictability and the associated privacy risks. It examines how aggregations of social contacts and non-social co-located individuals can provide substantial predictive information about a user's whereabouts, creating privacy leakage potential. Additionally, it considers perturbation and privacy-preserving mechanisms that incorporate semantic context to balance utility and privacy in location-based services.
Key finding: Demonstrates that social ties alone can provide up to 94% of an individual's mobility predictability, while aggregated non-social colocators can deliver up to 85%, indicating strong information flow from both sources. The... Read more
Key finding: Proposes a geo-indistinguishability-based perturbation mechanism that integrates location semantics and temporal consistency to generate plausible perturbed locations. This approach optimizes privacy protections by minimizing... Read more
Key finding: Develops a user-controllable, incremental Gaussian mixture model-based classifier for learning personalized location-sharing policies from user-labelled data. The method restricts policy evolution to local, interpretable... Read more