A Survey of Mobile Phone Sensing Michael Ruffing CS 495 Paper Info • Published in September 2010 • Dartmouth College – joint effort between graduate students and professors (Mobile Sensing Group) Outline • Current Mobile Phone Sensing – Hardware – Applications • Sensing Scale and Paradigms • Architectural Framework for discussing current issues and challenges Smartphone Technological Advances • • • • Cheap embedded sensors Open and programmable Each vendor offers an app store Mobile computing cloud for offloading services to backend servers iPhone 4 - Sensors Future Sensors • Barometer • Temperature • Humidity • To early to tell – cost and form factor will drive the availability of new sensors Applications • Transportation – Traffic conditions (MIT VTrack, Mobile Millennium Project) • Social Networking – Sensing Presence (Dartmouth’s CenceMe project) • Environmental Monitoring – Measuring pollution (UCLA’s PIER Project) • Health and Well Being – Promoting personal fitness (UbiFit Garden) Application Stores • Multiple vendors – Apple AppStore – Android Market – Microsoft Mobile Marketplace • Developers – – – – Startups Academia Small Research laboratories Individuals • Critical mass of users Application Stores • Current issues and challenges – User selection – Validation – Privacy of users – Scaling and data management Sensing Scale Sensing Scale • Personal Sensing – Generate data for the sole consumption of the user, not shared • Group Sensing – Individuals who participate in an application that collectively share a common goal, concern, or interest • Community Sensing – Large-scale data collection, analysis, and sharing for the good of the community Sensing Paradigms • Opportunistic Sensing - data collection is fully automated with no user interaction – Lowers burden placed on the user – Technically hard to build – people underutilized – Phone context problem • Participatory Sensing - user actively engages in the data collection activity – Supports complex operations – Quality of data dependent on participants Mobile Phone Sensing Architecture • Goal – architectural model for discussion • Components – Sense – Learn – Inform, Share, Persuasion Sense • Programmability – Mixed API and OS support for low-level sensors – Difficult to port application to multiple vendors • Continuous Sensing – Resource demanding – Low energy algorithms – Trade-off between accuracy and energy cost • Phone Context – Dynamic environments – Super-sampling using nearby phones Learn: Interpreting Sensor Data (Human Behavior) • Current applications are very much people centric • Learning algorithms – fits a model to classes (behavior) – Supervised – data is hand labeled – Semi-supervised– some of the data is labeled – Unsupervised– none of the data is labeled • Inferring human behavior via Sensors – GPS – Microphone Scaling Models • Scalability Key: Generalized design techniques that take into count large communities (millions of people) • Models must be adaptive and incorporate people into the process • Exploit social networks (community guided learning) to improve data classification and solutions • Challenges: – Common machine learning toolkits – Large-scale public data sets – Research sharing and collaboration Inform, Share, and Persuasion • Sharing – – – – • Visualization of the inferred data Formation of communities around the sensing application and data Community awareness Social networks Personalized Sensing – Voice recognition – Profile user preferences – Personalized recommendations • Persuasion – Persuasive technology – systems that provide tailored feedback with the goal of changing user’s behavior – Motivation to change human behavior • • • Games Competitions Goal setting – Interdisciplinary research combining behavioral and social psychology with computer science Privacy • Respecting the privacy of the user is the most fundamental responsibility of a phone sensing system • Current Solutions – Cryptography – Privacy-preserving data mining – Processing data locally versus cloud services – Group sensing applications is based on user membership and/or trust relationships Privacy – Current Challenges • Reconstruction type attacks – Reverse engineering collected data to obtain invasive information • Second Hand Smoke Problem – How can the privacy of third parties be effectively protected when other people wearing sensors are nearby? – How can mismatched privacy policies be managed when two different people are close enough to each other for their sensors to collect information? • Stronger techniques for protecting people’s privacy are needed Conclusion • Infrastructure has been established • Technical Barrier – How to perform privacy-sensitive and resourcesensitive reasoning with dynamic data, while providing useful and effective feedback to users? • Future – Micro and macroscopic views of individuals, communities, and societies – Converging solutions relating to social networking, health, and energy