Transcript Slides
Suman Nath Microsoft Research Contextual Computing Make computing context-aware • Context: location, activity, preference, history • A lot of progresses in location-aware services Not enough … • Need to use other signals o o o o o Do I like Italian restaurant? Am I walking? Do I drive 10 miles to eat? Is it lunch time or dinner time? Alone with family? Queue time ? Personal preference/history User’s context Real-time status • How do we get them? o Ask users to release more contextual information o Rely on crowdsourcing • Challenges to address: o Energy: partially solved o Privacy: mostly unsolved Energy • Many services require continuous sensing • Acquiring context is expensive • Many optimizations proposed o Not sufficient for continuous sensing o Phone will die in a few hours • Challenge: continuous sensing for a day without charging • Needs innovation: Efficient “Assisted” GPS Low Power Assisted GPS • Not regular GPS replacement 1ms NMS Takes 1s to minutes Same for ~150KM code phase Requires a few ms • Mobile phone sends to server: • • • Code phases Cell tower ID Time stamp • Server: • • Computes NMS Computes mobile location • Location-based services (e.g. mobile search) • Batched location estimation (e.g. path prediction) • Delay-tolerance positioning (e.g. geotagging photos) • Crowdsourcing LEAP: A Low Energy Assisted GPS for Trajectory-Based Services, Ramos et al. Ubicomp 2011 Privacy: do we care? • News: iPhone keeps record of everywhere you go Do people care? 48% 52% said they were "very or extremely concerned" about loss of privacy from using location-sharing applications Are you worried about geolocation privacy? 48% seriously concerned, 32% little worried Why is the stake high? Apple fined 1M won ($932) by South Korea over iPhone tracking allegations The suit now counts 26,691 plaintiffs => $26 million Lawmakers Demand Apple Clarify iPhone Tracking Capability Facebook fights new California privacy bill 'Do Not Track Me Online' privacy bill introduced by California Rep. Jackie Speier PER Theorem Revenue/ Relevance Privacy Efficiency Impossible to maximize all three Trivial to maximize any two Michaela Goetz and Suman Nath, Privacy-Aware Personalization for Mobile Advertising, no. MSR-TR-2011-92, August 2011 My wishlist • My context-aware service knows what is relevant • Without affecting my phone battery much • Without me telling it much about my private context • Even if I release limited private information o My privacy is preserved (even with strong adversaries) o In future I can revoke my data o (Only) I can decide how my data is used and shared