Transcript SERENDIPITI Platform Planning (City)

SERENDIPITI
SEnsoR ENricheD
Information Prediction and
InTegratIon
1
Serendipity n.
/ˌsɛ.rɛn.ˈdɪp.ə.ti
/
Scoperta (Italian)
Serendipiteit (Dutch)
Vrozené štěstí (Czech)
the effect by which one accidentally
discovers something fortunate,
especially while looking for something
entirely unrelated. (wikipedia.org)
2
Today’s Agenda
Vision
Objectives
Technology
Partners
3
Vision
What happens when you aggregate partial observations?
• Maps and ship captains
– No single human has been to every
point on a map
– Cartographers resolved partial
observations from ship captains
– Many needed, potentially conflicting
– Slowly, there emerged a map of the
world
• Can we do something similar to
learn something new about our
cities?
4
Objectives
• Aim? Stimulate and integrate research groups from
currently fragmented research areas, creating synergies
at the cross-over points.
• Approach? By fostering long-term relationships between
research groups, based around people, and laying the
foundations for a Virtual Centre of Excellence (VCE).
• In what areas? Real-time, large-scale data analysis and
inference for fusing semantic information and predicting
events.
• How? By harvesting, mining, correlating and clustering
extremely large, highly dynamic, very noisy, contradictory
and incomplete information from multiple sources
including: tweets, logs, RSS, web-sources, mobile texts,
web-cams, CCTV and other publicly-available multimedia
archives.
5
What’s New in SERENDIPITI?
• Real-time
• Prediction (events)
• Multimodal
• Noisy, errorsome data
• Novel applications
6
How do we do this?
• Sensing
– Aggregate diverse sources of information from the
real and online worlds
• Analysis
– Extract stable spatio-temporal patterns of human
activity
– Track these over time
• Use Case Scenarios
– City Planning
– Journalist (e.g. see Appendix)
– Police
7
How do we do this?
Physical
SensorBase
Online
8
How do we do this?
Physical
- Crowd movements, CCTV
- Traffic - pedestrian / vehicular
- Bluetooth sensing & proximity
- Blogs, wikis, web feeds
- Tweets, RSS
- Event guides
SensorBase
Online
9
How do we do this?
Physical
Track evolution of
events over space
and time
- Crowd movements, CCTV
- Traffic - pedestrian / vehicular
- Bluetooth sensing & proximity
- Blogs, wikis, web feeds
- Tweets, RSS
- Event guides
Online
SensorBase
Machine
learning and
data mining
10
How do we do this?
Physical
Track evolution of
events over space
and time
- Crowd movements, CCTV
- Traffic - pedestrian / vehicular
- Bluetooth sensing & proximity
- Blogs, wikis, web feeds
- Tweets, RSS
- Event guides
Online
SensorBase
Machine
learning and
data mining
SERENDIPITI
Platform
11
How do we do this?
Physical
Track evolution of
events over space
and time
Use Case Scenarios
Planning (City)
- Crowd movements, CCTV
- Traffic - pedestrian / vehicular
- Bluetooth sensing & proximity
- Blogs, wikis, web feeds
- Tweets, RSS
- Event guides
Online
Journalist
SensorBase
Police
Machine
learning and
data mining
SERENDIPITI
Platform
12
Partners & Roles
0 ERCIM
Patricia Ho-Hune
1 DCU-CLARITY
Alan Smeaton, Noel O’Connor, Barry Smyth
2 DERI
Giovanni Tummarello, John Breslin,
Paul Buitelaar
3 GLA
Keith van Rijsbergen, Joemon Jose,
Mark Girolami
4 QMUL
Ebroul Izquierdo
5 UvA
Maarten de Rijke, Arnold Smeulders
6 UEP
Vojtech Svatek
13
Partners & Roles
0 ERCIM
EU Project Management
1 DCU-CLARITY
Analysis of multimodal sensor data, real-time integration
of physical & online sources, organisation & management
of online sources
2 DERI
Semantic text analysis/mining, large-scale semantic
search/indexing, linked data, social semantics, online
communities
3 GLA
Multimedia information retrieval, formal models, mining
information from large data sets, event detection,
information fusion, machine learning
4 QMUL
Correlating/mining media and textual data, sensor
base, automatic (CCTV-based) event analysis
5 UvA
Focused crawling, wrapper induction, mining social
media, information extraction, data integration, crossmedia mining and information fusion, machine learning
6 UEP
Mining rich associations from large databases,
information extraction from the Web, semantic and social
14
Web technology, effectiveness of ICT
PARTNERS
1
DCU
2
DERI
Physical
3
GLA
4
QMUL
Track evolution of
events over space
and time
5
UvA
6
UEP
Use Case Scenarios
Planning (City)
- Crowd movements, CCTV
- Traffic - pedestrian / vehicular
- Bluetooth sensing & proximity
- Blogs, wikis, web feeds
- Tweets, RSS
- Event guides
Online
Journalist
SensorBase
Police
Machine
learning and
data mining
SERENDIPITI
Platform
15
PARTNERS
1
Physical
DCU
2
DERI
1 4
3
- Bluetooth sensing & proximity
Online
5 1
2 6
GLA
4
QMUL
Track evolution of
events over space
and time
- Crowd movements, CCTV
- Traffic - pedestrian / vehicular
5
- Blogs, wikis, web feeds
- Tweets, RSS
- Event guides
3
5
UvA
UEP
Use Case Scenarios
3
2
6
4 SensorBase
6
5
3
Machine
learning and
data mining
6
SERENDIPITI
Platform
1
Planning (City)
2
3
Journalist
4
5
Police
6
16
Work Packages
• WP1 – Management
• WP2 – Integration of Organisation & People
Joint Program of Activities
• WP3 – Real-time Large-scale Data Analysis
• WP4 – Semantic Information Fusion
• WP5 – Inference & Event Prediction
• WP6 – Applications & Infrastructure Sharing
• WP7 – Outreach (Spreading Excellence)
17
Implementation
User Group
Planning (City)
Journalist
Police
Data Provision Board
Wikimedia
Foundation
Boards.ie
TW
S+V
EPA/MI/Met
Industrial
Advisory Board
Ken Wood
Milek Bover
E. Aarts
Carto Ratti
One other !
18
Tangible Outputs
• Provision of mini projects to tackle
unforeseen research topics
• Industrial placements of research staff
• Academic exchanges
• Contribution to standardisation efforts
• Public software repositories and
tools/data access through the
SERENDIPITI platform
• Outreach to other targeted projects
19
Impact
• SERENDIPITI :
“SEnsoR ENricheD Information Prediction and InTegratIon”
• People - Research - Outreach - Platform
• Large-scale, semantic urban computing
20