Harnessing the Power of Data Science Through Research Chair: John Pullinger National Statistician Professor Andrew Blake Director, The Alan Turing Institute
Harnessing the Power of Data Science Through Research
Chair:
John PullingerNational Statistician
Professor Andrew BlakeDirector, The Alan Turing Institute
The Alan Turing Institute 2
Professor Andrew BlakeInstitute Director
The Alan Turing Institute
3
Harnessing the power of Data Science through research
The Alan Turing Institute
The UK’s National Institute for Data Science
‘We will found the Alan Turing Institute to ensure Britain leads the way again in the use of big data and algorithm research’
George OsborneBudget Speech, March 2014
The Alan Turing Institute
The data economy – principal players
01/05/2023
5
a decreasing number of people will gather to themselves the knowledge that comes from owning giant databases
– The Telegraph on Jaron Lanier’s “Who owns the future?”
The Alan Turing Institute
... and powering growth for mid-size companies
01/05/2023
6
The Alan Turing Institute
... and creating value via smaller companies
01/05/2023
7
$5bn$3bn
~ $500m
~ $100m
~ $100m
The Alan Turing Institute
An Institute without disciplinary boundaries
The Alan Turing Institute
Interdisciplinary science
The Alan Turing Institute
Priorities for research
10
The Alan Turing Institute
Working with government – relationship with ONSCollaboration between Turing Fellows Suzy Moat, Tobias Preis and the ONS Data Science Campus
Barchiesi, Moat, Alis, Bishop and Preis (2015)
“Quantifying International Travel Flows Using Flickr”, PLOS ONE 10.
A map of the world built only from GPS locations of Flickr photos
The Alan Turing Institute
HSBC programme in data-driven economics
12
The Alan Turing Institute
The HSBC – Turing Strategic Partnership ...
The UK’s National Institute for Data Science
“… we are excited about the prospect of working with the Institute’s world leading scientists using big data analytics to better understand economic trends.”- Douglas Flint, HSBC Chairman
... aims to help economists, researchers, policymakers and businesses to better understand the UK economy and its interconnection with global markets.
www.turing.ac.uk
The Alan Turing Institute
Resilient networks
14
The Alan Turing Institute
Evaluation of Velocity Fields via Sparse Bus Probe Data in Urban Areas
Richard Gibbens and Andrei Bejan, Computer Laboratory, University of Cambridge, 2015
The Alan Turing InstituteRichard Gibbens and Andrei Bejan, Computer Laboratory, University of Cambridge, 2015
10am, Tuesday 15th June 2010 10am, Tuesday 22nd June 2010
Urban traffic velocity fields via “bus probe”
The Alan Turing Institute
Machine Learning meets mathematics
17
dd
The Alan Turing Institute
Rough paths to Chinese characters
The Alan Turing Institute
Rough Paths to Chinese Chars
The Alan Turing Institute
AI for data analytics
20
The Alan Turing Institute
... AI for data analytics
Zoubin Ghahramani
James R. Lloyd
The Alan Turing Institute
Empirical classifier– black box, optimize predictive power, efficient
Generative model– explain observations, Bayesian inference
? Which wins
Machine learning:– big data or explanatory models?
The Alan Turing Institute
Example: online recommender system
MovieRecommender
• Generative model – hidden variables
• Performance engineered – predictive black box
eg Netflix challenge
The Alan Turing Institute
Black box (RBM)
(eg Salakhutdinov et al. 2007)
Y
X
W
User
Movie
𝑝 (𝑌∨𝑋 ;𝑊 )
The Alan Turing Institute
Generative model
(eg Stern et al. 2009)
The Alan Turing Institute
Trait space
The Alan Turing Institute
Deep neural networks
(Krizhevsky, Sutskever, Hinton, 2012)
(LeCun, Y., Bottou, L., Bengio, Y. & Haffner, P., 1998)
The Alan Turing Institute
Deep neural network for speech classification
© Microsoft Research
The Alan Turing Institute
Captioning an image
Microsoft team
Also teams from Google, Stanford, ..
The Alan Turing Institute
Transparency Fairness
Full value from data
Generative models: set to make a comeback?
The Alan Turing Institute
Fast mapping (Bloom, 2000)
The Alan Turing Institute
Tufas and other objects learned from very few labels
The Alan Turing Institute
“Fast mapping” with logic
The Alan Turing Institute
Learning: the data efficiency trade-off
??
1 10 100 1000
Tolerance to noise
No of examples
?
The Alan Turing Institute
Black box recognisers
Generative models Transparency and fairness
Efficiency of learning – logic?
The Alan Turing Institute
The Turing’s strategic priorities
36
#econstatsJoin the conversation: