Upload
hoangdang
View
227
Download
0
Embed Size (px)
Citation preview
http://mobilityresearchcenter.org © 2009 Carnegie Mellon CyLab. All rights reserved.
ARCHITECTURE
Intelligent Meeting OrganizerArrange Meetings QuicklyAlejandro Rivera, Kathleen Yang, Neha Pattan, Martin Griss
PROBLEM• Meeting scheduling
– with multiple participants• Decision factors
– Changing participants schedules– Optimal meeting venue– Inflexible room resources (video, teleconference) – Various personal preferences
SOLUTION• Virtual Secretary
– Learns your personal preferences– Manages your daily agenda– Negotiates meetings for you– Re-schedules meetings– Knows your location, recommends a suitable venue– Reminds you of scheduled meetings
• Users:– Doctors, Patients, Anyone!
Mobile ClientPhilosophy:
• “Request it and forget it”
Advantages:
• Always with you wherever you go
• Offline capabilities
• Easy to useFunctionality:
• Location-aware: GPS & GSM Cell ID
• Offline capabilities
Environment:
• Cell phoneXMPP
HTTP
SMS
Communication protocols
SecretaryPhilosophy:
• “Arrange it and notify user”Environment:
• Personal Computer (laptop)
• Company’s central serverAdvantages:
• Computing power
• Access to personal data
• Privacy: “Don’t share personal information”Functionality:
• Communicates with other secretaries
• Negotiate meetings
• Tracks scheduled meetings
STATUS• Arrange meetings with multiple participants• Get meeting parameters from user and mobile
sensors• Negotiate meetings between secretaries based on
users agenda (Google Services)
TECHNOLOGIES• Mobile client:
– Python for Symbian S60– XMPP / HTTP / SMS– Google Maps service
• Server side:– Java Agents Development
Framework (JADE)– Google Calendar/Contacts
services– XMPP
Secretary Negotiation
User Virtual Secretary Calendar
NEXT STEPS• Add indoor location capabilities.• Indicate attendees location on a map prior to the
meeting.• Predict user behaviors based on historic data• Introduce advanced user preferences that will feed
the rule system used by the secretary to negotiate meetings.
• Create different types of secretaries and behaviors to fit different types of users.
Sponsors: Cylab, Nokia
© 2009 Carnegie Mellon CyLab. All rights reserved.http://mobilityresearchcenter.org
Usable Security and Privacy for Context-Aware Mobile Applications
How the medical world can benefit?
Diwakar Goel, Eisha Kher, Shriya Joag , Martin Griss, Anind Dey
GOAL
Improve end-users’ ability to privately and securely interact with their environment and data while mobile.
PROBLEM
METHODOLOGY
Camera in conjunction with GPS and cell-id :
1.User travels to a region.2.Receives a QR Code valid for a limited time based on his location.3.Uses this token along with QR codes (static or dynamic) at service site.
• To access a region• To access sensitive information
4.Authenticates (event registered with server)• Logged on the server
. Expires code as per lifetime
CHALLENGES
• Manual focus required• Move processing to server?
• Have to pause to capture• Varying phone screen resolutions
BENEFITS
• No other hardware needed• Camera available on most phones• Screens are everywhere!
• Inexpensive to generate• Can be changed on the fly• Revoking access easier• Easier for administration
• Minimizes human errors in security• One time setup!
Interacting with a smart environment challenges users: • how do you provide input? • how do you perceive output?• what can the system do for you?• how will it do it for you?…
Mobile users face particular challenges as they move:• realizing differences between the environments• discerning what they can and cannot do and what data they may have access to
SOLUTION
Making the various sensors on the phone and in the environment:• mutually negotiate and authenticate each other• to access services, gain privileges and take implicit actions.
•Accelerometer•Camera•Microphone•GPS
•Bluetooth•Wi-Fi•GPRS/3G/EDGE•Cell-ID
•Battery•IMEI•Memory•SW Version•IMEI
Calendar, Email, Call / SMS history
© 2009 Carnegie Mellon CyLab. All rights reserved.http://mobilityresearchcenter.org
IMSafeLeveraging IMS Innovation™ and Android through a Mobile Health Application
Alain Kajangwe, Chandani Desai, Diwakar Goel, Patrick Maniraho Kristoffer Gronowski, Martin Svensson
PROBLEMHow do you:
Constantly stay aware of medical facilities nearby? Communicate crucial medical information to paramedics when unconscious? Effectively locate the closest person in your safety net? Get information from other IP devices when you cannot
access your phone? Keep track of your medication and lab appointments?
SOLUTION
Intelligent location-based services using Google Maps Securely store crucial medical information directly on your phone for ease of access Store – forward – broadcast presence information through IMS Innovation servers Multimedia synchronization through IMS
IP Multimedia Subsystem Not Just Text Powerful ways of forking Being adopted by mobile operators Presence notification Supports subscribe – notify (push) Better security architecture Messages can have different priority levels XDMS (XML Document Management Server) XCAP (XML Configuration Access Protocol)
IMS:fills the gap
ANDROID
Open Source Interesting libraries and constant updates to APIs Designed to work with next-generation mobile technologies Easy interfacing with 3rd party sensors and devices
FEATURES
FUTURE WORK:
Real time inventory status from pharmacies nearby Dedicated device to trigger emergency actions Interface additional condition sensors with to phone Use sensors and robust communication technologies for diagnosis
Enhance native contact list with IMS presence Fallback to other communication mechanisms e.g. SMS Synchronization with other multimedia devices Emergency actions Medication and lab appointment scheduler Medicine inventory management
http://mlt.sv.cmu.edu © 2009 Carnegie Mellon CyLab. All rights reserved.
Mobile Natural Language Technologies
Joy Ying Zhang, CMU [email protected]
Mobile Speech Translator• Pandora: Two-way statistical speech
translation system for mobile devices (PDA, cellphone)
• Data-driven: easy to train for new language pairs and domains
• Domain: travel/medical/force protection• Languages:
English/Spanish/Chinese/Japanese/Thai ...
Mobile Sign Translator Automatically detect text area in captured image
Optical character recognition (OCR): picture to text
OCR error correction as a translation task Apply machine translation on foreign text
On-going Research•Context-aware speech translation (translate according to where you are, who you are)
•Self-adapting translation system
ReferencesZhang and Vogel, Pandora: a large-scale two-way statistical machine translation system for hand-held devices, MT Summit XI, 2007Zhang and Huang, Mining translations of OOV terms from the web through cross-lingual query expansion, SIGIR 2005.Chang, Zhang, Vogel and Yang, Enhancing image-based Arabic document translation using a noisy channel correction model, MT Summit XI, 2007Yang, Gao and Zhang, Towards automatic sign translation, HLT 2001.
Speech Translator for 3D Virtual World
• Multilingual speech conversation in the virtual 3D world (Second Life)
• Simulates what users will do in real life• Automatic language detection• Automatic translation
Capturer Module
Interactive Module
Camera Image
Image Commands
Commands
Results
Visual Output
Detection & Recognition &
Translation Module User
Audio Output
User Input
http://mobilityresearchcenter.org © 2009 Carnegie Mellon CyLab. All rights reserved.
PROBLEM• GSM doesn’t work well indoors• Zero-setup is better• Fewer sample fingerprints should work• Fingerprint pollution by signal fluctuation
Enhancement of Wi-Fi Indoor LocationingTony Lin, Ilya Landa, Martin Griss, Joy Zhang
SOLUTION• Indoor locationing technique• Using existing Wi-Fi access points (AP)• Enhanced Redpin algorithm is introduced• Dominant AP filter is adopted
FUTURE WORK• Tracking with user activities• More RF sources in addition to Wi-Fi,
such as BlueTooth• More pilots applications
CONCLUSION• Enhanced Redpin can provide
efficient accuracy even with fewer training data
• Applying the dominant AP filter can gain better results by reducing noise
USER SCENARIOS• Tracking medical personnel in a hospital
– To study personnel movement in a hospital– To find the nearest relevant person
• Tracking elders at home or senior center– To determine current location of a person– To view movement history– To detect abnormalities in a person’s position or movement
EXPERIMENT• Five algorithms were used to do comparison
– Naïve Bayes Classifier– Support Vector Machine (SVM) – K Nearest Neighbor (KNN)– Redpin– Enhanced Redpin
• Different training data size was measured• Adjacent areas were measured• Applying AP filter to reduce the signal noise• Different number of APs was measured• 1,000 fingerprints were used• 10% for testing data, 90% for training data• For each experiment, 100 runs were executed
Bayes SVM KNN Redpin Enhanced Redpin
Accuracy 61% 81% 80% 86% 87%
Accuracy after apply filters
10% 90% 89% 90% 91%
Figure 1: Average accuracy of non-filtering and filtering data for each algorithm
Figure 2: Average accuracy for every training data size
Figure 3: Average accuracy of adjacent areas Figure 4: Average accuracy of different number of APs
0.4
0.45
0.5
0.55
0.6
0.65
0.7
0.75
0.8
0.85
0.9
50 100 150 200 250 300 350 400 450 500 550 600 650 700 750 800 850 900
Size of training data
Acc
urac
y
Naïve BayesSVMKNNRedpinPMI Redpin
0.50.55
0.60.65
0.70.75
0.80.85
0.90.95
1
0 1 2 3The number of increased adjacent rooms
Acc
urac
y
Naïve BayesSVMKNNRedpinPMI Redpin
0.6
0.65
0.7
0.75
0.8
0.85
0.9
0.95
1 2 3 4 5 6 7 8
The number of access points
Acc
ura
cy
SVMKNNRedpinPMI Redpin
Sponsor: CyLab
© 2009 Carnegie Mellon CyLab. All rights reserved. http://mobilityresearchcenter.org
Social ShopAds
HealthDine Measurement
Problemv Provide monitoring, feedback, and encouragement for the user to comply with clinician or care giver-set goals.
v Sensors on mobile devices collect and analyze data to determine user activities.
v Monitor for compliance and give feedback to the user.
v Provide summaries and notifications to the clinician or caregiver.
v Ensure all communications are secure.
v Ensure user has control over what information is given to whom.
v Ensure security and privacy should not be over burdening on either the mobile device or the user.
Status
v Built a prototype on N95 using AES and on a serverv Initial rule-based privacy model
• Privacy Rules Definition: Through web portal• User Mediation: Through device, application created
for device, accessed through Wi-Fi .• Artificial Intelligence: Learning done by server, based
on rules and user mediation.v Evaluated several security schemes
• Measure efficiency of different encryption schemes.• Authenticating users of the system• Generic curve of data size versus CPU usage• Ensuring security of the keys, creation and distribution
of new keys
Next Stepsv Explore different learning schemes
v Test prototype in multi-user campus setting
v Location Tracking to provide better medical data
v More efficient power consumption
v Study of comparison of individual privacy rules with the social group
v Scalability of system for large number of sensors and large quantities of data being sent
http://mobilityresearchcenter.org
Solution & Benefits
v To determine the most optimal solution for encryption of datav Create user-defined privacy specificationsv A machine learning paradigm that learns from the previous decisions of the usersv Measuring overheads for the transmission of the data and make the process more efficientv Ensuring a reliable system in all media with respect to wireless communication of the data over large distances.v Key infrastructure element in remotely supervised, home- based patient rehabilitationv Better and less expensive alternative for rehabilitation.v No supervised rehabilitation or monitoringv Early detection of abnormal conditions
Sponsor: CyLab
Real Time Health Monitoring SystemSecurity and Privacy
Sweta Deivanayagam, Anusha Nagarajan, Yash Shroff, Ashish TulsainiMartin Griss, Asim Smailagic, Dan Siewiorek
© 2009 Carnegie Mellon CyLab. All rights reserved.http://mobilityresearchcenter.org
A Mobile Application To Monitor the ElderlyOmar Abdul Baki, Tony Lin, Martin Griss, Joy Zhang
ProblemCurrent solutions which monitor the homebound elderly are either too costly or too intrusive.
Solution1. An “always-on” mobile application2. Learns behavior patterns3. Responds to erratic readings 4. Classifies activities according to :
a) Locationb) Movement and posture
Scalable Accurate Non-Intrusive Low-Cost Usable
Preliminary Results
Prototype in Mobile Python for the Nokia N95 :• Uses online clustering for learning• Detects falls• Recognizes repeated behavior as normal• Detects movement from room to room• Detects changes in postures (standing,
sitting, lying down)
Future Work
• Factor time of day and day of week into learned behavior patterns
• Use other available sensors (such as light and microphone)
• Replace thresholding with more advanced abnormality recognition techniques
Prompt: Are you all right?
Continue Monitorning
YES NO
Sponsors: CyLab, Panasonic, Nokia
© 2009 Carnegie Mellon CyLab. All rights reserved.http://mobilityresearchcenter.org
Personal Messaging AssistantProvide focused and timely healthcare information messagesDeepthi Madamanchi, Sumalatha Komarraju, Senaka Buthpitiya & Martin Griss
ProblemHealthcare providers interact with many peoplethroughout the day. Currently many support peopleprocess messages, reminders and follow-ups. • Increasing use of mobile devices and mail by
patients and caregivers around the clock• Clinics and hospitals offering email and web
access services to patients• Support personnel have limited working hours • Increasing delays in providing medical help in
emergency and critical conditions.• Possible loss of business and eventual distrust
of a particular HMO or healthcare provider.• Providers work in multiple hospitals with
different policies and modes of patient interaction How PMA works
Next Steps• Handle more message types and sources
• Process forward, file, folder, calendar rules
• Enhance intelligence and statistical message content analysis
• Deploy PMA on a single secure personal service center
• Larger scale test and pilot
• Add features for other mobile professionals
SolutionPartially automate processing of message traffic to mobile devices. • An intelligent, context and content-aware
messaging system which can act more effectively/rapidly than a human assistant.
• “Triage” health-related email content to match context, importance, and urgency.
• Deliver the right message at the right place and right time based on context and preference of the healthcare provider or patient.
Scenario• Dr. Bhatt uses e-mail to follow up on his cardiac
patients’ post-release from hospital, sends drug orders to pharmacies, and replies to (some of) his patients’ queries.
• These tasks have different levels of urgency and importance.
• PMA can direct the most appropriate message(s) to the doctor based on his current context (location and activity) or to his nurse or colleagues.
Features of PMA• Context Aware
Processes e-mail based on your location and activity into consideration
• IntelligentWorks out importance and urgency of emails (based on your context and preferences)
• AdaptiveLearns your priorities and preferences
• ExtensibleSupports new contexts (e.g., provider signs up for a weekly medical conference)
Results• Functional prototype using Jess rules, and
MySQL on server, accessing email from Gmail inbox
• Sends selected prioritized messages via SMS (or HTTP) to N95 phone
• Accepts user context & tags, and GPS location
Sponsor: CyLab
Motivation Emergencies like earthquakes, may have
individuals trapped in unsafe buildings or spaces.
Limited manpower and risk to the lives of emergency response workers impede rescue efforts.
Traditional static sensor networks, need impractical number of nodes to cover large areas simultaneously.
Robots have high per unit cost. They can be deployed only in limited numbers thus offering lower coverage.
SensorFlyCollaboratively-mobile Miniature Flying Sensor Network
Aveek Purohit and Pei Zhang
Maximum access to unknown areas Cannot rely on infrastructure Self recovery from failures
Challenges
Platform
Mapped Safe Area
Base/Charging Station
Replacement
Low Power
SensorFly: Overview
SensorFly Middleware Architecture
First Generation SensorFly: Low-cost compass sensor to determine the
orientation Accelerometer to detect collisions with
obstacles Sonic sensor for altitude detection V1 SensorFly node weight = 29 grams Energy usage for V1 SensorFly node
Future Work: Flight control and navigation with minimal sensing. Energy efficient network protocols for the controlled mobile SensorFly environment. Collaborative 3D localization
© 2009 Carnegie Mellon Cylab. All Rights Reserved.http://mobilityresearchcenter.org
Motors
Acc
Compass
Sonic
Radio
Altitude Control
Navigation
Scheduler Network ManagementTask
Management
Applications
Hardware / Peripheral Drivers
Location Engine
Operation Mode Power UsageHovering 6.1 WForward flight 7.2 WData Tx/Rx/Sensing 310/330/225 mWProcessing only 150 mWIdle ~1 mW
Sponsor: Cylab
http://www.cmu.edu/silicon-valley/research/smartspaces.html © 2009 Carnegie Mellon CyLab. All rights reserved.
SmartSpaces Mobile Health and Telehealth CapabilitiesMartin Griss (PI), Ray Bareiss, Patricia Collins, Ed Katz, Mike Smith
Indicator 1 – Number of Older Americans
• 81% prefer to live independently• 30% would “rather die” than live in a nursing home• Limited mobility • Declining health• Home health expenditures in the U.S. are less than in other developed countries
Elders need assistance at home
Mobile blood pressure monitor
• Tracks vital signs
• Notifies family/friend, health care provider of abnormalities
• Notification ranges settable by family/friend, health care provider
Exercise monitor
• Tracks motion throughout the home (RFID, Wi-Fi)
• Reminds patient of need to exercise
• Optionally, provides exercise instructions
• Notifies family/friend of patient’s failure to move
• Notification information is settable by recipient
Diet monitor
• Tracks selection of grocery products
• Reminds patient of dietary restrictions
• Optionally, provides nutrition information
• Notifies family/friend of patient’s failure to comply
• Notification settable by recipient
Home HomePortal
Home HomePortal
Home HomePortal
ServiceCenter
ConciergeService
ServiceService
ServiceService
ServiceService
ServiceService
Analytics
Family & Friends
Concierge
Caregivers
Ecosystem Architecture(future work)
Context-Aware, Agent-Based System
• Supports aggregation of information across homes
• Supports initiative-taking agents
• Supports autonomous collaboration by agents
• Concierge mediates some services to optimize health care provider’s time
• Family & friends offer first-line support to patient when appropriate
Solution• Integrated, context-aware, agent- based, mobile system• Multiple modalities (voice, touch, vision, RFID)
Results
Sponsors: SAP, Panasonic, Nokia, Motorola