2013-10-10 robust and trusted crowd-sourcing and crowd-tasking in the future internet
- 1. “ENVIROfying” the Future Internet
THE ENVIRONMENTAL OBSERVATION WEB
FOR THE CROSS-DOMAIN FI-PPP APPLICATIONS
Robust and trusted crowd-sourcing and crowd-
tasking in the Future Internet
ISESS 2013, Oct. 09-11 2013
Denis Havlik, Maria Egly, Hermann Huber, Peter Kutschera, Markus
Falgenhauer, Markus Cizek (all AIT Austrian Institute of Technology GmbH.)
- 3. Copyright © ENVIROFI Project Consortium 3
Enviromatics
meet Future Internet
Future Internet
• Networking technology
• Infrastructure as a Service
• Internet of Things, Content,
People
INSPIRE, GMES, SISE
• Geospatial
• Environmental Observations
• Model Web, Sensor Web,
• Data Fusion, Uncertainty
ENVIROFI
FI-PPP
Environmental Usage Area
• FI Requirements
• Specific Enablers
• Envirofied cross-area Applications
- 4. ENVIROFI Scenarios
1. Bringing Biodiversity into the Future Internet
• Enabled biodiversity surveys with advanced ontologies
• Analysis, quality assurance and dissemination of
biodiversity data
1. Personal Information System for Air Pollutants,
allergens and meteorological conditions
• Enhance human to environment interaction
• Atmospheric conditions and pollution in “the palm of
your hand”
1. Collaborative Usage of Marine Data Assets
• Assess needs of key marine user communities
• Selection of representative marine use cases for further
trial: leisure and tourism, ocean energy devices,
aquaculture, oil spill alert
Copyright © 2013 ENVIROFI Project Consortium 4
- 6. Challenge: human sensors
Copyright © 2013 Denis Havlik, AIT Austrian Institute of Technology GmbH.
Illustration by Scoobay (http://www.flickr.com/photos/scoobay/224565711/)
6
- 8. Challenge: plausibility and QA
8
Image Classifier SE
Quality Assessment SE
classify & check images
Image Archive SE
manage images
MDAF server
mobile acquisition
General User
leaf images
metadata (e.g. geotag)
Expert
leaf species
manual assessment
© 2013 ENVIROFI Project Consortium
- 9. Microlearning helps!
Copyright © 2013 Denis Havlik, AIT Austrian Institute of Technology GmbH. 9
Objective Possible approach
How to use the application? Tooltips or popup messages
on first use (implemented)
Training to recognise
objects
Scavenger hunt for known
and tagged objects
Learn to avoid
misidentifications
control questions &
feedback
A-posteriori feedback Notify user when more info
on the object is available
(implemented)
Classify data & assess
users knowledge
Generalized re-capcha
principle
Microlearning in a sense of „learning while doing“ is crucial for quality of
information but still not sufficiently taken into account. On TODO list.
- 10. Observation DB
Challenges: no single truth
10
Plausibility/Confidence checks
Consensus building
Previous situation
knowledge
Habitat
Informatio
n
Image
Recognition
Reporters
Reputation
Observ. on things
(independent,
conflicting,
incomplete)
Observations on
observations
(identification,
plausibility, annotation)
Application
specific views
(fusion, meaning
uncertainty)
Sensor
Networks
ENVIROFI
observations
ENVIROFI
observations
Integrate
existing data
Integrate
existing data
USE
Copyright © 2013 Denis Havlik, AIT Austrian Institute of Technology GmbH.
- 11. Challenge: FI-Ware integration
11
„Cloud Edge“ GE:
Field-deployment of observations server;
(P2P?) information exchange over local
WLAN
„Cloud Edge“ GE:
Field-deployment of observations server;
(P2P?) information exchange over local
WLAN
11
Observations &
Situation
Awareness
Cloud Storage
Storing of BLOBS
(photos, videos)
Cloud Storage
Storing of BLOBS
(photos, videos)
Marketplace GEs:
sales, revenue
sharing
Marketplace GEs:
sales, revenue
sharing
Pub/sub GE:
Events processing
& dissemination
Pub/sub GE:
Events processing
& dissemination
Security GEs:
user & right mgm.;
legal compliance
Security GEs:
user & right mgm.;
legal compliance
IoT GEs:
Smart sensors?
IoT GEs:
Smart sensors?
Environmental SEs
Meaning
Data Fusion,
Forecasting
Harvestors,
Connectors
Observations
Big data GEs:
Annotation &
processing
Big data GEs:
Annotation &
processing
Cloud mgm. GEs:
automated
deployment, scaling
Cloud mgm. GEs:
automated
deployment, scaling
I2ND GEs:
Network reliability,
Hardware abstraction,
I2ND GEs:
Network reliability,
Hardware abstraction,
Mashup GE:
Ad-hoc applications
Mashup GE:
Ad-hoc applications
Copyright © 2013 Denis Havlik, AIT Austrian Institute of Technology GmbH.
- 13. Challenges: user interaction
View existing knowledge
•Map view
•Table view
•Detailed View
•Areas of Interest
View existing knowledge
•Map view
•Table view
•Detailed View
•Areas of Interest
Receive information (events!)
•Requests for more observations,
•Warnings, e.g. “pollen warning”
•Interests, e.g. “monumental tree in
vicinity”
Receive information (events!)
•Requests for more observations,
•Warnings, e.g. “pollen warning”
•Interests, e.g. “monumental tree in
vicinity”
Report observations
•“New” things, e.g. “here and now I
see a tree”
•Personal, e.g. “I have a headache”
•Obs. on existing thing, e.g. “this
tree currently blossoms
Report observations
•“New” things, e.g. “here and now I
see a tree”
•Personal, e.g. “I have a headache”
•Obs. on existing thing, e.g. “this
tree currently blossoms
Inform
Server
Backend
(or proxy)
Alert!
Request
Action!
Copyright © 2013 Denis Havlik, AIT Austrian Institute of Technology GmbH. 13
- 15. Challenges: tracking of users?
Work with „Areas of Interest“:
•defined by users
•Could be automatically generated
when user significantly moves
Why AOIs?
1.pre-fetching of data => allows
offline use
2.Server-side filtering of events
No tracking!
Current users position is taken into
account by the Mobile app logic!
Copyright © 2013 Denis Havlik, AIT Austrian Institute of Technology GmbH. 15
1 2
3
4 5
6
7 8
AOIAOI
AOIAOI
AOIAOI
- 16. Do it on the phone!
Copyright © 2013 Denis Havlik, AIT Austrian Institute of Technology GmbH. 16
Do you really need to
process users sensitive
data (e.g. health-
related) on the backend
server?
- 19. Copyright © 2013 Denis Havlik, AIT Austrian Institute of Technology GmbH. 19
Lessons learned?
- 22. 1. The ideas presented today were developed and
partially realized as Mobile Data Acquisition
System (MDAF) in the scope of the European
Community's Seventh Framework Programme
(FP7/2007-2013) under Grant Agreement Number
284898 (ENVIROFI)
2. The importance of microlearning really became
clear to me very recently, thanks to Dr. Christian
Voigt and the microlearning 7.0 conference.
3. MDAF contributors: Eun Yu, Clemens Bernhard
Geyer, Peter Kutschera, Markus Falgenhauer,
Markus Cizek, Ralf Vamosi, Maria Egly, Hermann
Huber and most recently Jan von Oort.
• Currently active developers are underlined.
Acknowledgements
22
- 23. • All slides which are marked as © 2013 Denis Havlik
or © 2013 Maria Egly can be re-used under the
terms of the Creative Commons ”Attribution-
ShareAlike 3.0“ license.
• Illustration on the pages 2 and 6 have been marked
for free re-use by their authors. To the best of my
knowledge the licenses are compatible with CC.
• Disclaimer: I am not a lawyer. Please follow the links
for more info.
• The logos on slides 17 and 23 are of course IPR of
the respective companies. To the best of my
knowledge this falls into “fair use”
IPR and fair re-use
23
- 24. Thank you for your attention
Dr. Denis Havlik
denis.havlik@ait.ac.at
The research leading to these results has received funding from the European Community's Seventh
Framework Programme (FP7/2007-2013) under Grant Agreement Number 284898
www.envirofi.eu
Editor's Notes
- Two out of three ENVIROFI scenarios are strongly biased towards citizen scientists, mobile crowdsourcing and crowdtasking and local situation awareness.
- In some senses, humans are „bad sensors“. They are non-standardized, difficult to calibrate, don‘t like the idea of working 24/7, easily bored and their accuracy and sensitivity erratically varies over time. However, they also excell at pattern recognition and interpretation of the results.
This makes them complementary to hardware sensors and very valuable for some types of applications.
- Unlike standard monitoring systems, the human sensors (and to a lesser extent also information from user-owned sensors) inevitably deliver conflicting and incomplete information. The Quality assurance of such data often relies on combination of peer review, expert opinions and various indicators.
In MDAF, all these results can happily co-exist even if they are contradicting each other. The decision „what is the reality?“ is only made at the level of „application specific view“, taking into account the owners interests and trust in various data sources.
As a result, it is perfectly possible to generate several conflicting „realities“ from the same data set. E.g. a Greanpeace applicaiton will show different reality than a fisherman association applicaiton simply because they make different assumprions concerning the relative importance and trustworthiness of the data.
The „applicaiton specific view“ has not been fully implemented, but the technology is the same as the one used for the quality assurance part.
- The architecture shown here goes beyond the ENVIROFI pilots and indicates our ideas what woudl be possible to do with FI-Ware GEs in the future.
- This demo has been developed within the project scope and brought to working PoC status. We are confident that the concept will work well when we start making „real“ applicaitons. In fact, we (AIT) are using this in CRISMA project now, so the number of available widgets and our know-how steadily rises…
- Tasking of volunteers and experts is a key to data collection and quality assurance. It is crucial to task the users which are both able and willing to perform this task, while avoiding the information overflow.
In ENVIROFI, AIT was able to develop a concept and basic technology which will allow us to implement the context- and profile- specific tasking in the future projects.
- I‘m not sure which is the license for the slides which I „inherited“ here, sorry. I‘m sure that *I* can use them, and I presume the right to re-use them will be granted to anyone who asks. Please contact the respective consortium leaders.