Weitere ähnliche Inhalte Ähnlich wie Smart Cities UK 2018- Stream 1 Data (20) Mehr von Scott Buckler (20) Kürzlich hochgeladen (20) Smart Cities UK 2018- Stream 1 Data3. How A.I. can influence UK
Cities and Towns - Speaker:
Vishal Chatrath, CEO and
co-founder, PROWLER.io
5. © PROWLER.io 2018 – www.prowler.io
In AI/machine learning
perception/classification
is considered solved
6. © PROWLER.io 2018 – www.prowler.io
The next big challenge: decision-making AI
AI that can make transparent & reliable decisions in
complex, dynamic, uncertain & high-dimensional
environments in real-time
7. © PROWLER.io 2018 – www.prowler.io
Decision-support tools
Complexity timeline
inputcomplexity
80s 90s 2000s 10s70s60s1950s
expert
systems
10. © PROWLER.io 2018 – www.prowler.io
Deep Neural Nets: effective,
automated perception
DNNs are, a very efficient tool for recognition / perception
tasks:
● Are not data efficient. Use blind repetitive speed
instead of probability
● Can't handle complex behaviour with high
dimensionality (bounded environments)
● Employ a single, narrow learning technique
● Can’t handle uncertainty in the environment
Tool for classification
/perception tasks
PROWLER.io: adaptive, data efficient
AI systems for decision-making
PROWLER.io has created a transparent, dynamic decision
system that:
● Can learn and adapt as the environment evolves
● Can handle complex behaviour with high dimensionality
● Can work in a multi-agent environment
● Can handle uncertainty in the environment
Principled AI
11. © PROWLER.io 2018 – www.prowler.io
Probabilistic modelling
Allows us to build a continuous probabilistic model of a
environment. Models can manage high dimensional factors influencing
the environment with a measure of uncertainty.
Reinforcement learning
Enables agents to plan and evolve strategies in complex goal
settings. The behaviour of the world is typically not deterministic, but
probabilistic. So RL needs probabilistic approaches to make good
decisions.
Multi-agent systems
Realistic scenarios have dynamic environments. Environments
dynamically change as agents react to each other’s actions. An
intelligent agent must model the behaviour of other agents in order to
understand their intentions and why they react the way they do.
Our unique combination
Three distinct branches of maths in one platform
12. © PROWLER.io 2018 – www.prowler.io
● Every decision-making node is an autonomous AI agent.
● Each agent can sense and dynamically adapt to high-dimensional
inputs with 1000s of parameters.
● Inputs can be information about the environment or incentives, which
are used to automatically tune behaviour and interaction.
● Any actions by agents in the environment update the model of the
environment dynamically.
● Mechanism Design (Inverse Game Theory) enables the design of
incentives that steer agents toward desirable outcomes, producing
overall outcomes with the desired risk characteristics.
How do we do it?
Allow me to explain...
13. © PROWLER.io 2018 – www.prowler.io
Platformcomponents
GamesFinance Logistics
Tunable AI
Bounded rationality
GP State Space Model
Resource allocation
Demand matching
Mechanism design
Hierarchical RL
Risk-sensitive RL
Spatio-temporal model
Probabilistic
modelling
Reinforcement
learning
Multi-agent
systems
Autonomous
systems
Decision-support
simulations
Deep GP
Model-based RL
Platform
Deployable Decision-Making System
Cloud Learning System
Our platform provides an AI toolbox
Decision-making across multiple industries
15. © PROWLER.io 2018 – www.prowler.io
Demand
moving set of objects
to be served by agents
Supply
A fleet of agents
Supply & Demand
Distributions over space
Objective:
Minimise cost - difference between
distributions over time horizon and movement
penalty.
19. © PROWLER.io 2018 – www.prowler.io
Resource allocation
Modelling the future
● Probabilistic models predict
the demand for tomorrow
● A series of possible tomorrows are
simulated (“Monte Carlo” samples)
We can evaluate the effect of taking decisions
● For each possible “tomorrow”,
we solve the resource allocation
problem in simulation
● See the effects of decisions
(e.g. number of trucks)
20. © PROWLER.io 2018 – www.prowler.io
Resource allocation
Heterogeneous resources & tasks
Different
assignment
tasks
21. © PROWLER.io 2018 – www.prowler.io
…
t2t1 tn
Long-term plans
Dynamic tasks & resources
22. © PROWLER.io 2018 – www.prowler.io
Demand matching
& resource allocation
Dynamic allocation with
varying resource due to
drop-off
23. © PROWLER.io 2018 – www.prowler.io
Conceptual structure of a system
Transportation example
Model of environment
● Form a continuous probabilistic
model of city environment
● Models can manage high complexity
of factors influencing the
environment
Training of agents
● Set targets at agent or system level
● Decide if agents collaborate and if they
impact environment with actions
● Find optimal strategies for agents
dependent on current and predicted
states of system
● One or many agents
Build a well coordinated
team of experts to
support decision making
● Operate in the real world using the
agents as decision support
or autonomously given the
latest state
24. © PROWLER.io 2018 – www.prowler.io
Model of environment
● Form a continuous probabilistic
model of market environment
● Models can manage high complexity
of factors influencing the
environment
Training of agents
● Set targets at agent or system level
● Decide if agents collaborate and if they
impact environment with actions
● Find optimal strategies for agents
dependent on current and predicted
states of system
● One or many agents
Build a well coordinated
team of experts to
support decision making
● Operate in the real world using the
agents as decision support
or autonomously given the
latest state
Conceptual structure of a system
Financial example
25. © PROWLER.io 2018 – www.prowler.io
Model of environment
● Form a continuous probabilistic
model of game environment
● Models can manage high complexity
of factors influencing the
environment
Training of agents
● Set targets at agent or system level
● Decide if agents collaborate and if they
impact environment with actions
● Find optimal strategies for agents
dependent on current and predicted
states of system
● One or many agents
Build a well coordinated
team of experts to
support decision making
● Operate in the real world using the
agents as decision support
or autonomously given the
latest state
Conceptual structure of a system
Games example
26. © PROWLER.io 2018 – www.prowler.io
PROWLER.io
Adaptive, data efficient, highly complex applications
Application specific
Dataefficiency
Application complexity
27. © PROWLER.io 2018 – www.prowler.io
• Our principled AI can make reliable decisions in complex, dynamic & high-
dimensional environments under uncertainty in real-time.
• We have moved AI from being a black box to something that is transparent reliable,
& controllable.
This is the world’s first principled AI platform
that generalises decision-making
PROWLER.io
Summary
28. © PROWLER.io 2018 – www.prowler.io
Financing:
• Seed round of £1.5M closed in Aug 2016
• Series A investment round of £10M closed in Jul 2017
Leading independent AI company:
• 20 papers in 12 months by the team to date
• UK, Cambridge based; 60+ full time staff including 29 Ph.Ds
• World leading researchers in the areas of Probabilistic Modelling, Game
Theory and Reinforcement Learning have been attracted to join PROWLER.io’s journey
• Attracting talent from around the world - 24 nationalities and counting
Commercial traction across key verticals:
• Financial institutions, logistics, smart-cities, games
PROWLER.io
Today
29. The power of Big Data to
improve social and economic
outcomes - Speaker: George
Johnston, Founder, Nitrous
30. The power of big data to improve social and
economic outcomes
1 February 2018
32. 2020: ⅓ of
procurement
from SMEs
Public sector innovation
2025: £20
billion market
Talent,
capital and
policy
B2G2C
Consumers =
citizens
34. Civtech
Service and data innovation
Service delivery improvements
Govtech
Back office, procurement data
innovation
Efficiencies and cost savings
Govtech and civtech
36. Data driven approach
Open data platform
Platform
products
Central government
Local government
Private sector
Data scientists
SMEs
Talent
CityMapper of X?
38. Mapping - 5G deployment:
think global, act local -
Speakers: Ordnance Survey,
Met Office and the University
of Surrey
39. Research and real world
planning for 5G networks
Think global act local
• FI FT H GENERAT I O N MOBI LE COMMU NI C AT I ON S
Richard Woodling
February 2018
© Ordnance Survey 2018
40. Today…
• Why 5G?
• The Bournemouth experience
• Building the 3D digital model
• Building a propagation model
• Interactive planning tool
• Real world planning
© Ordnance Survey 2018
41. Why 5G?
“By 2023, over 30 billion connected devices1 are forecast, of which around
20 billion will be related to the IoT. Connected IoT devices include
connected cars, machines, meters, sensors, point-of-sale terminals,
consumer electronics2 and wearables. Between 2017 and 2023, connected
IoT devices are expected to increase at a CAGR of 19 percent, driven by
new use cases and affordability.” Ericsson IoT predictions
• 1,000x increase in capacity
• Support for 100+ billion
connections
• Up to 10Gbit/s speeds
• Below 1ms latency
© Ordnance Survey 2018
42. The DCMS Project ask of geospatial
• To commission research into
geospatial considerations for
network planning:
• Brings together appropriate
geospatial and other data
for the siting of high
frequency (mmWave) radio
infrastructure
• Identifies the location of
optimum sites for radio
equipment to provide
efficient and effective
capacity coverage
Considerations:
• Frequency ranges above 6GHz (
specifically the IMT-2020 bands,
including 26GHz, 32GHz, 39GHz
and 60GHz)
• Vegetation, buildings, walls,
weather, other obstacles
• Process of validating the output
data
and verification in the field
• Implications, including those to
other countries
© Ordnance Survey 2018
44. • Three detailed study zones
– dense urban
– suburban
– rural
• Rich civic data
• Rich 3D data
Bournemouth study zone
© Ordnance Survey 2018
46. Additional assets
© Ordnance Survey 2018
Street furniture
o2000 lamp posts
o4028 sign posts
o2926 traffic lights
o292 park benches
o1987 bus shelters
Telecoms features
o1152 telephone boxes
o262 internet fibre
boxes
o365 grey boxes
o274 green boxes
o349 telephone
cabinets
o 30+ data sets
o 18+ tools
o 5.6Bn points in the cloud
o 2,700 Km of underground assets;
high and low voltage cable routes,
ducting and CCTV networks
o 2,500 Aerial images
o 87 bridge extent polygons
47. Location: positional accuracy
• Real world 3D point
cloud
• Optimised
infrastructure
• Efficient and effective
coverage
Supplied
location
Actual
location
OS data resolution 10cm
© Ordnance Survey 2018
49. Modelling matters for planning
“I can simply deploy loads of antennas (sites) – can’t I?”
• Signal to interference and noise ratio (SINR)
• Knife and shield diffraction
• Resolving received power
• Network initialisation (Base stations deployment)
• Simulation and dynamic evaluation
• Mapping of SINR to Capacity using Modulation /
Coding lookup tables
• Coverage maps
© Ordnance Survey 2018
50. 5GIC – Measurements for the built
and natural environment
• Shield & edge diffraction
• Vegetation loss
• Scattering
• Material characteristics
• Field validation
© Ordnance Survey 2018
51. Weather impact modelling
T+0 UKPP @ 2 km T+0 UKPP @ 100 m
Stephan Havemann, Robert Scovell, Jean-Claude Thelin, Dave Jones
1. Select significant weather events from the archive
2. Downscale to “5G resolution” (100m)
3. Calculate the attenuation
Extinction
@ 26 32 42 71 84 GHz
© Ordnance Survey 2018
52. The interactive
demonstrator tool
• Demonstrates how built and natural environment
including weather, impacts signal propagation
• Geared to planners and business users
• Built in core features for planners
• Incorporate user parameters
• Scalable
© Ordnance Survey 2018
56. Planning in the real world
(what matters?)
• Where?
• What’s the demand/service (use case)?
• Dynamic considerations
• Location geo-spatial challenges
© Ordnance Survey 2018
57. Candidate locations for mmwave
• Urban streets and pedestrian
areas
• Transport hubs
• Retail complexes
• Large event sites and stadia
• Business districts
• Dense residential areas
• Small town environments
• Road and Rail networks
© Ordnance Survey 2018
58. The role of Geo-spatial for planning
© Ordnance Survey 2018
Example - Urban Streets and pedestrian areas
60. What do I capture and how?
© Ordnance Survey 2018© Ordnance Survey 2018
• Area of coverage
• Scheduling (when)
• Resolution (spec.)
• Budget / cost
• Time to survey (how long)
• Access constraints
10cm 40cmor
61. Additional data sets
• Footfall in a shopping centre
• Individual retail unit customer
counts
• Car park usage
• Transport timetables and
ticketing information
• Major events planning
• Ticketing for events
• Professional surveys of people
movement such as that provided
by Health and Safety Labs (HSL)
• Others as available
• Currency - when was it last
updated and what is the update
period and method?
• Provenance – Can you trust the
source of this data? Who
created it and how?
• Accuracy – Does it provide
figures commensurate with the
intended use?
© Ordnance Survey 2018
62. Conclusions
• A detailed 3D geospatial model is essential for accurate
modelling of coverage and capacity
• Multiple data sources are required, and much baseline data exists
• Quality of data can be variable
• Maintaining a 3D neutrally held geospatial model is non-trivial
• Use cases need to be clearly articulated
• Planning tools are essential
© Ordnance Survey 2018
Detail matters – A global challenge needing a local view
63. Where next for OS?
• Exploring spatial enhancements to mapping
• Refining planning tool capability
• Working closely with Central and Local Government
© Ordnance Survey 2018
66. CHICAGO - MUNICH - NEUKIRCH - REDWOOD CITY
Intelligent Data for
Clever Parking Solutions in Cities
Chris Wortley
Director Sales Airports
1st. February 2018
Smart Cities UK
67. copyright Cleverciti Systems 2018
High search traffic
Air pollution
Threat by EU penalty fees
Non-payment for parking
Inefficient enforcement
Stressed cardrivers
Challenges for Cities
68. copyright Cleverciti Systems 2018
New data
& knowledge
Convenient
Parking
Traffic safety
& efficient
enforcement
Less costs,
More revenue
Reduction of
air pollution
Cleverciti‘s solutions provide …
69. copyright Cleverciti Systems 2017
Intelligent real-time parking
and traffic flow management
Focusing on outdoor parking
Highly efficient and patented
technology
Overhead sensors scanning
10 to 100 spaces per sensor
Easy-to-install system e.g.
at existing lamp posts
Full-service end-to-end solutions
or integration into existing data
platforms
Cleverciti‘s products
73. copyright Cleverciti Systems 2017
Using new or existing Apps
Optional: Reservation
Optional: Carfinder
1. Navigation from home to a free parking space
closest to your preferred destination
Mobile App
75. copyright Cleverciti Systems 2018
Winner Digital Innovation Award
DIGICON – Digital World Conference
Best Munich Startup
Winner Fast Mobility Category - Bits&Pretzels
Winner Innovation Award 'Deloitte Fast 50‘
Finalist Energy Awards 2016
Finalist Best New Parking Product
Smart Mobility Gulf Traffic Award, Dubai
Innovation Trail Parkex 2017
German Accelerator Program 2017
Ecosummit Award Berlin 2017 – Bronze
Winner Digital Cities - EIT Digital Award 2017
Cleverciti – Prices and Awards
76. copyright Cleverciti Systems 2018
30 Countries. 40 Installations.
Bad Hersfeld · London Westminster · Dubai · Dublin · Cologne · Munich · Berlin
Rotterdam · Lubljana · Vancouver · Caloundra, Australia · Chicago · Vienna Airport
Fuerth · Moscow · Florence · Boston · Gent · Weiden · Eindhoven · Geneva
Everywhere
77. copyright Cleverciti Systems 2018
Welcome...
www.cleverciti.com
Cleverciti System GmbH
Hofmannstrasse 54
81379 Munich, Germany
Cleverciti Systems Corp.
100 S Saunders 1st floor
Lake Forest,
IL 60045, USA
CLEVERCITI is a registered trademark
of CLEVERCITI SYSTEMS GMBH
Cleverciti Sensor Technology & Systems,
Cleverciti App Features, Cleverciti Circ CMS 360° are
internationally Patent Pending
78. Open data for Smart Cities
Speaker: Fanny
Goldschmidt,
OpenDataSoft