People-centric Pervasive Sensing: Enabler for a reality-driven Internet

Transcription

1 Human-centered Computing Lab People-centric Pervasive Sensing: Enabler for a reality-driven Internet Paulo Mendes (paulo.mendes@ulusofona.pt) Telefonica R&D Barcelona October 16 th,

2 Overview COPELABS: Overview From Wireless Sensing Networks to People-Centric Sensing People-centric Sensing: Properties, Applicability and Architectural Issues Paulo Mendes, PhD Vice-director COPELabs Director of PhD programme NEMPS Associated Professor University Lusofona Before: Scientific director SITILabs Senior Researcher NTT Docomo Research visitor Columbia University PerSense: A framework for Pervasive Sensing Focus on data sharing in pervasive environments

3 Overview COPELABS is a private, non-profit research association Development and dissemination of activities in the context of informatics sciences and of cognitive and mind sciences. Merges two University Lusófona R&D centers: SITILABS (Informatics) and CEPCA (Psychology) Being established in October 2013 as multidisciplinary R&D unit R&D track 1: Internet Science R&D track 2: Interactive Cognitive Psychology 27 researchers, 19 with a PhD Senior researcher, researcher, and junior researcher levels

4 Cognitive Behavior + Informatics: Why? Availability of open-source technology gave rise to people-centric services Internet stakeholders acknowledge the user as prosumer Pervasive technology can assist in creating new models (e.g. mhealth) People-centric approaches require cognitive understanding of technology impact User behavior (social, technological, emotional) has an impact on Internet design and services Reality-driven services require a deeper look into social aspects, to evolve with the society pace Costs of supporting cognitive awareness bring complexity to the overall system, increasing inequality GOAL: developing people-centric services, in the field of mhealth improving the citizen's daily lives Social interaction models by applying a multidisciplinary perspective Cognitive enhancement based on intensive technology Behavior inference and assessment to improve daily living

5 Research Topic Pervasive Wireless Systems Focus: Pervasive wireless systems, derived from social interaction and human behavior inference Coordinator: Paulo Mendes Main Aspects Connectivity and behaviour dynamics requires a self-organizing approach to new wireless design Focus on Pervasive People-centric Sensing Systems Research Goals: Self-awareness Advanced levels of autonomy to adjust continuous sensing to changing conditions Awareness based on behavior patterns in order to exploit phones spatial distribution and context diversity. Synchronization Synchronization as a requirement to cooperative sensing Development of algorithms to achieve sensing synchronization based on behaviour similarities Emergence and Network Dynamics Focus on human inspired models based on behaviour inference and social interaction Assess the emergence of social sensing systems from individual systems.

6 Research Topic Social Internet Design Focus: Evolution of the Internet as networking structure, based on social interaction and human behavior inference Coordinator: Rute Sofia Main Aspects Internet as social structure that generates knowledge: look at its social capital potential Prosumer role and behaviour requires a shift in the Internet architectural design Research Goals Open Data sensing and behavioral inference: Context characterization based on data from the networking perspective: social interaction; familiar stranger Roaming prediction as a applicability example Social Internet modeling Applicability of social capital modeling metrics (trust, influence) to evolve the Internet architectural design. Evaluate the networking impact that different interpretations of the same metric (centrality or betweenness) Software Defined Networking. Introduction of more flexibility on the networking layers Development of networking services based on user-centricity.

7 Research Topic Interactive Cognitive Psychology Focus: Cognitive enhancement by designing advanced ICT solutions focused on Interactive Cognition, e.g. seeking to empower individuals with mental and neurological impairments. Coordinator: Pedro Gamito Main Aspects To promote accessibility to treatment; To promote patient s and caregivers adherence to treatment; Self-adaptive, pervasive, collaborative and user-friendly systems Research Goals Increase cognitive functionality Immersion of patients in virtual reality worlds that were devised to boost cognitive strategies Study of disruptive behaviours and impairments Assessment of cognitive functionality Based on behavioral tasks, and tracking Evaluation of cognitive impairments in real environments Impact of introducing pervasive and mobile solutions to assist long-term treatments Novel tools to bring benefits to the individual, and society Assessment of real-time assessment and correction methodologies

8 Our Ecosystem A glimpse for the Internet Science perspective Simula Universities (Aveiro, Lisbon-IST, Coimbra) Nonius Software Caixa Mágica Software EFACEC Engenharia e Sistemas S.A. ZON YDreams Ambiesense University of Kent Alcatel-Lucent Bell Labs Sen.se Technicolor Huawei Technical University of Berlin DT-Labs Cisco UCLA (Prof. Lixia Zhang) University New York - Poly (Prof. Shivendra Panwar) University Federal of Pará CPqD Tecnalia FON Telecom Italia University of Urbino University of Rome

9 Wireless Sensing Systems Why aren t we looking at those? Application specific: not a significant impact in our daily lives Live people out of the loop. Lack of economy of scale We look at personal pervasive sensing devices: Smartphones Always with us Internet-enabled Locatable (GPS or other systems) Full of sensors Have economy of scale

10 People-centric Sensing Properties & Applicability Major properties Large-scale, long-lived, mostly mobile Application-agnostic Very limited multi-hop wireless Not so energy-constrained Mobility is a driving factor People is in the loop Security, trust and privacy are important factors Applicability Personal Sensing: Emotional Maps Sensing people behaviour Sensing individual emotions Sensing proximity interactions Sensing patterns of conversation Correlation of people behaviour and cognitive interactions Community Sensing: Tracking Disease Spreading Assess the evolution of disease spreading. Identifying who has been in contact with a potential infected person Know when and where the contact took place Know the distance and contact time with the suspect, the logical places for the meeting, the relationship with the suspect Group Sensing: Integrated Physical and virtual Networks Social networks do not allow people to know where their friends are and what they are doing in real-time Private information should be located in personal devices Information made available to friends will also include behaviour inferred by the group. 10

11 People-centric Sensing Mobile Sensing Architecture Little or no consensus about a mobile sensing architecture Robust context classifiers Common methods for collecting and sharing data Smart scheduling of sensing commitments Not clear solution for positioning of architectural components Example: raw sensor data should not be pushed to the cloud because of privacy issues. Components: Sense: mobile phones collect raw sensor data from sensors Learn: Information is extracted from the sensor data by applying machine learning and data mining techniques (phone and/or cloud) Inform, share, persuasion: Web sensing: sharing sensorial information on the web Mobile Computing: ability to compute and mine big data from very large numbers of users

12 People-centric Sensing PerSense: Pervasive Sensing Framework Goals Modular sensing and data processing framework for mobile and embedded devices Computing, communication, and sensing anytime, anywhere Sensing and Tracking Inference Pervasive Data sharing Call log Bluetooth CitySense Pervasive Sensing Framework Opportunistic Computing WiFi GPS Accelerometer PerSense Pervasive Sensing Framework Pervasive Sensing PerSense Pervasive Sensing Framework Big Data Analysis Data Centric Opportunistic Networking PerSense Pervasive Sensing Framework Mobility Tracking Online Social Networks Post 12

13 People-centric Sensing PerSense Platform: Open Source Code We are developing a variety of libraries that allow developers to quickly build applications that leverage smartphone sensor data, trigger context-based notifications, and collect survey responses. General Information Developed in C # / Mono (Monodroid, Monotouch). Mono is a software platform designed to allow developers to easily create cross platform application. Mono is an open source implementation of Microsoft's.NET Framework based on the ECMA standards for C# and the Common Language Runtime. Available for: Windows Mobile, iphone, Android.NET micro framework for embedded devices (soon). Code available Maestroo: Get it on Google code <soon> Mtracker: Get it on Google code <soon> ICON: Get it on Google code <soon> The libraries are released under the GPLv3 License. Get in touch if you are interested in finding out more. Siti-mgt@siti1.ulusofona.pt Sensing and Tracking Maestroo TM MTracker TM Inference Spin-off SenCeption Pervasive Data Sharing ICON TM

14 People-centric Sensing Pervasive Data Sharing General Information Based on the Named Data Networking concept Aims to improve performance on opportunistic networks Main technical innovation: Content based social-aware opportunistic routing Self-organized in-network caching Software Design Features Modular design: Decision, Network, Data Engines Declarative Decision Engine Dependency injection software design pattern (TinyIoS) Component selection at run-time Multi-treading architecture Can be configured and updated remotely

15 Pervasive Data Sharing Social-aware Content-based Opportunistic Forwarding Motivation Scenarios with intermittent connectivity even in urban environments (e.g. high cost, fading, closed APs), but: power-constrained devices while increasing storage capability majority of applications are related to data gathering Data exchange in challenge networking scenarios performance improvement by exploiting social interactions and structure users are not interested in knowing the location of data SCORP: Social-aware Content-based Opportunistic Routing Content knowledge (i.e., content type, interested parties) Social proximity ê Faster, better content reachability in challenged networks

16 Pervasive Data Sharing SCORP: Utility Functions & Algorithm Probability of encountering nodes with a certain content interest among the ones with similar daily social habits CD(a,d3) CD(a,b1) CD(a,b1) CD(a,b1) CD(a,e4) CD(a,b1) CD(a,c2) CD(a,b1) CD(a,c2) CD(a,f1) CD(a,c2) 08:00 a.m. 12:00 p.m. 04:00 p.m. 08:00 p.m. 12:00 a.m. 04:00 a.m. 08:00 a.m. Daily Sample T i A W(a,1)i W(a,2)i 1 2 A 1 2 A 1 2 A 1 2C A 1 2 A Time-Evolving Contact to Interest x :

17 Pervasive Data Sharing SCORP: Evaluation Scenario Opportunistic Network Environment (ONE) Synthetic mobility model: 12-day interaction in the city of Helsinki 150 nodes divided into 8 groups of people and 9 groups of vehicles. Human mobility traces (Crawdad): 36 nodes in Cambridge University 2 months Traffic 32 sporadic contacts per hour Message size: from 1 to 100 KB. Buffer space: 2 MB Time-To-Live (TTL): 1, 2, 4 days, 1, and 3 weeks Benchmarks dlife: Bubble Rap: 24 daily samples of one hour K-clique for community detection Spray and Wait (serves as lower bound in what concerns delivery cost): L = 10 Results analysed in terms of: Average delivery probability: ratio between delivered messages and messages that should have been delivered Average cost: number of replicas per delivered message Average latency: time elapsed between message creation and delivery

18 Pervasive Data Sharing SCORP: Network Load Impact based on traces #of replicas % Average Delivery Probability (1-day TTL) Spray and Wait Bubble Rap dlife SCORP # of messages/interests per node Spray and Wait Bubble Rap dlife SCORP Average Cost (1-day TTL) # of messages/interests per node Seconds Major Findings Average Latency (1-day TTL) Spray and Wait Bubble Rap dlife SCORP # of messages/interests per node Spray and Wait has low performance (nodes follow routines and do not cover the whole simulated area) and low cost/high latency (few nodes are used to forward). Bubble Rap has low performance due to buffer exhaustion; highest cost for 1msg/int due to community creation. dlife and SCORP have similar behaviour: dlife may lead to buffer exhaustion (approximately 24% more than the allowed) SCORP: keeps resource usage at a low usage rate. Up to 93.61%, 90.25% and 89.94% less latency than Spray and Wait, Bubble Rap and dlife

19 People-Centric Sensing Enabler for a reality-driven Internet We envision that Future applications will: be developed based on a realistic view of society allow a more natural fit into peoples needs. Such real-time adjustment of applications to the way people live and interact in society will be possible by correlating: different sources of open data and crowd-source data with context data sensed from peoples behaviour and social environment (social data).

20