Data Science: Exploring the

Mathematical Foundations

November 2014
Data Science: Exploring the Mathematical Foundations

Contents
1. Executive Summary ..................................................................................................................... 3
2. Introduction ................................................................................................................................ 4
3. Background ................................................................................................................................. 4
4. Opportunities and Motivations................................................................................................... 4
4.1. Mathematical and Statistical Techniques ........................................................................... 5
4.2. Big Data ............................................................................................................................... 6
4.3. Reducing and Understanding Uncertainty .......................................................................... 6
4.4. Bridging the Gap between Industry and Academia ............................................................ 7
4.5. Cross-Sector Knowledge Transfer ....................................................................................... 7
5. Challenges ................................................................................................................................... 7
5.1. Big Data ............................................................................................................................... 8
5.2. Communication ................................................................................................................... 8
5.3. Validation and Verification ................................................................................................. 9
5.4. Skills Shortage ..................................................................................................................... 9
5.5. Data Security and Anonymity ........................................................................................... 10
6. Strategic Priorities ..................................................................................................................... 10
6.1. Data Science Community .................................................................................................. 10
6.2. Conferences and Workshops ............................................................................................ 11
6.3. Education .......................................................................................................................... 11
6.4. Encouraging Mathematical Activities ............................................................................... 11
7. Recommendations .................................................................................................................... 12
8. Acknowledgements................................................................................................................... 12
9. References ................................................................................................................................ 12
Appendix A - Workshop Agenda ....................................................................................................... 14
Appendix B – Delegate List ............................................................................................................... 15

2
Data Science: Exploring the Mathematical Foundations

1. Executive Summary
This report provides a summary of the opportunities, challenges and potential future strategies for
mathematics and statistics within data science, and the added value that the mathematical sciences
can bring to industry. It is based on the findings from a joint workshop on the application of the
mathematical sciences to the underpinning foundations of data science, held by the Knowledge
Transfer Network and the Smith Institute. The workshop was attended by a mix of academia and
industry, giving a variety of perspectives from across a range of sectors, all with one thing in
common – they want to get more out of data. Domains represented at the workshop included digital
forensics, social media, environment, engineering, technology, security, defence and aerospace.

There has been an upsurge of commercial and academic interest in data science. It has become a
crucial tool to handle, manipulate and analyse data on which many important decisions are based.
Across all sectors of industry and academia, it is recognised that adding new streams of data or
finding patterns in existing data can add value to business. With the ever increasing amount of data
available, the role of data science becomes ever more important and the mathematical sciences are
a key element for its success.

The table below summaries the opportunities, challenges and strategic priorities, including
suggested future activities and noted key principles, for mathematics and statistics to add value to
business within the area of data science:

Opportunities Challenges Suggested Activities Key Principles

extracting maximum analysis gives KTN Ltd and emphasise co-development

value from data actionable and InnovateUK to help of solutions between data
compatible options coordinate leadership users, analysts and providers

providing proof of balancing security, improve access to high develop mechanisms that
concept to privacy and quality data kick-start opportunities and
demonstrate value accessibility of data new ideas

formalising, ensuring analysis incentivise encourage commitment of

quantifying and and visualisation collaboration between industrialists to exploit
refining hypotheses are dynamic industry and academia opportunities and impact

capitalising on the access to high encourage open promote the sharing of

advances made in quality data and research (results with experiences, good practice
computation expertise data provided) and data across domains

creating a data finding both data prevent intellectual property

science community science and sector or legal requirements from
specific expertise inhibiting progress

Each point listed in the above table is discussed in more detail within the report.

3
Data Science: Exploring the Mathematical Foundations

2. Introduction
The Knowledge Transfer Network (KTN) and the Smith Institute (SI) held a joint workshop on the
application of the mathematical sciences to the underpinning foundations of data science on 23rd
July 2014. The workshop was attended by delegates who were specially chosen for their wide-
ranging and relevant expertise as well as their appreciation of the mathematical sciences and data
science.

This report captures views and opinions that the delegates expressed during the workshop to
determine the major mathematical challenges that academia and industry face within data science
practice in the UK. Strategic activities and recommendations are given to overcome the challenges in
data science and thus increase added value to industry.

3. Background
Industry is continually facing the challenges of dealing with large, complex and sometimes fast-
moving data sets that are difficult to process and learn from. There is demand from UK business to
know more about different types of data analysis, the methods involved and techniques applied. The
demand is driven from the added value that using data more intelligently might bring. However, it is
all too easy to misunderstand the data and its structure or to apply inappropriate analytical
techniques which consequently draw flawed conclusions. The evolving field of data science is
therefore gaining prominence.

Data science is the field of study which is concerned with the collection, preparation, analysis,
visualisation and management of data. It is an area which builds upon and incorporates the expertise
of many different disciplines to successfully extract meaning from data.

Whilst the computing infrastructure to store and handle data is a necessity, mathematical sciences
are fundamental in underpinning the ideas, concepts and techniques required to analyse data and
ultimately extract the useful insights that allow industry to make decisions that create growth.

The current position of the UK with respect to data is fast-moving. There are more data about the
way we live than ever before and the volumes of data are continuing to increase. There is an
increase in the deployment of sensors to track information across a vast range of areas from
transport to banking, retailing to farming, science to energy. Social media alone is providing
remarkable insights into customer behaviour and emotional preferences. Given the current
situation, the mathematical science community have a great opportunity to support UK business.

4. Opportunities and Motivations

The following benefits can be achieved through appropriate analysis of data: new scientific
discoveries, market-changing products, increased transparency, improved decision making and
enhanced services. Couple significant volumes of data with the benefits of good data analysis and
industry, academia and government have the opportunity to enhance the competitiveness of the UK
and grow the economy. The UK is particularly well placed to do this because it has a very strong
mathematics and statistics community.

The opportunities for the mathematical sciences to add value to business are:

4
Data Science: Exploring the Mathematical Foundations

 applying appropriate mathematical and statistical techniques to extract maximum value

from data;
 providing proof of concept for new technical areas to demonstrate methods’ value to
industry;
 formalising and quantifying hypotheses about how the data arose and validate, compare
and refine those hypotheses;
 capitalising on the advances made in computation which allow greater flexibility and choice
of mathematical and statistical methods;
 creating a data science community, rich in mathematical and statistical knowledge, that can
boost the economic growth of UK companies.

The above opportunities as well as examples of areas that were identified during the workshop are
described in more detail below.

4.1. Mathematical and Statistical Techniques

Amongst the workshop delegates there was a feeling that many industries are suffering a data
deluge. The data collected are often complex and the information to be extracted multifaceted;
simple counting methods are insufficient to assemble the required information or produce only
simple answers. There is a shift towards the use of more advanced and adaptable methods that have
the ability to process structured or unstructured data, of changing sizes and dimensions, that is
discrete or continuous, and with varying levels of complexity.

There is a wide variety of mathematical and statistical techniques and complex analytics that can
extract value from complicated, multifaceted data. Throughout the workshop there was much
emphasis on the successful use of
New areas of study – Topological Data Analysis
probabilistic methods and the need to (TDA) example
disseminate them more widely whilst
Many mathematical and statistical techniques are
encouraging understanding by a wider
emerging which on paper provide numerous benefits to
community. Other techniques are listed data science. As these methods are, however, new to the
below with references for further reading data science scene there is a limited number of
gathered at the end of the report: applications and case studies of their use.

TDA for example is a new area of study aimed at having

Probability Theory [1] applications in areas such as data mining. TDA
Hidden Markov Models [2] represents data using topological networks and uses data
Evolving and Multiplex Networks [3,4] sampled from an idealised space or shape to infer
Deep Learning [5,6,7] [13]
information about it . It essentially allows algorithms to
Bayesian Analysis [8] analyse sets of data to reveal the inherent patterns within
Classification and Clustering [9,10] rather than showing correlations between preselected
Data mining [11] variables.
Graphical Models [12] This method could be valuable to industry because it
Topological Data Analysis [13] enables the user to discover insight into the data without
Tropical Geometry [14,15] having to ask the correct question of the data in advance.
Dynamical systems [16] By working with industrial partners to have access to real
Machine learning [17] data sets, case studies can be published to demonstrate
Sparse Tensor Methods [18] value and promote new techniques to industry.
Stochastic optimisation tools [19]
Large-scale Linear Algebra [20]

5
Data Science: Exploring the Mathematical Foundations

Not only can these methods be used to extract value from data but they can be used to judge the
performance of competing data analytics tools, in terms of computational efficiency, accuracy,
stability and scalability to larger problems.

4.2. Big Data

Big data is generally understood to refer to techniques developed to analyse data sets which are
either too big, too complex or too lacking in structure to be analysed using standard approaches. A
common misconception around big data is the expectation that acquiring powerful computer
infrastructure will immediately provide a business advantage. Instead information technology,
computer science and mathematical science must go hand in hand. Infrastructure is necessary, but
achieving value from big data also
Big Data Opportunities
requires more sophisticated data
With the right analytical and strategic approach, big data has analysis methods.
the potential to be of great social benefit by unleashing the
power to tailor services, thus enhancing user experience and New approaches to analysing data
consumer satisfaction. The combined strength of data can lead
have to be found and, where
to enriched information, enabling the extraction of ground truths
and the mapping out of a ‘complete picture’. For example, appropriate, existing methods have to
tracing a drug from its creation to the end of trials and beyond be scaled. This is where the
allows for more advanced mathematical modelling of disease mathematical sciences can make a
patterns and progression, which in turn will provide a better
understanding of individual variability, improved risk and impact
considerable contribution: building on
assessment, and enhanced decision support. the foundations of current statistical
methods and identifying new
Further examples of big data opportunities for the mathematical
sciences lie hand in hand with the fact we are living more and techniques to augment or replace old
more in a service-oriented era. Smart cities and intelligent ones that are less appropriate, making
systems, such as buses tweeting service times and delays, or the analytics efficient, and most
home heating systems texting temperature alerts for instance,
importantly making sure the correct
are already important application areas. The design and
development of mathematical algorithms which provide the inferences are drawn from the data
intelligence for these systems is a key component. available.
Customer facing industries are data rich. Retail industries with
For big data analytics to make the
their loyalty card schemes, utility companies with the
implementation of smart meters to monitor homes, and phone greatest impact the mathematical
companies with the rise of smartphones with GPS and internet sciences need to apply data analytics
connectivity are just three examples out of many which have to areas where the UK has competitive
growing markets on a global scale and an abundance of data.
advantage. If there is a real need and a
real problem, then this is where true value can be gained.

4.3. Reducing and Understanding Uncertainty

Uncertainty exists in many aspects of data science and there is a need to better understand and
reduce this uncertainty. Mathematics and statistics can formalise and quantify hypotheses about
how the data arose and validate, compare and refine those hypotheses. In this way, we can get an
understanding of the underlying “laws of motion” and also make predictions and explore what-if
scenarios.

Perhaps more widespread is the opportunity to introduce probabilistic thinking, such that implicit
uncertainty can be communicated, visualised and interpreted more accurately. Even with high
volumes of data and the best analytical techniques, uncertainty cannot be eliminated. It is therefore

6
Data Science: Exploring the Mathematical Foundations

important that we can measure the amount and understand the types of uncertainty that surround
us in everyday life.

4.4. Bridging the Gap between Industry and Academia

The workshop delegates represented a mix of academia and industry and through discussions it was
apparent that there is a need to bring the two closer together. There is enthusiasm to establish
stronger relationships and create a data science community which covers research and commercial
domains spanning many sectors.

As an example, customer-facing industries have potential for huge growth. With a natural shift
occurring to become more customer-focused, mathematical sciences can further pave the way to
help understand what customers want, and how to analyse large quantities of data to extract value
and gain business and customer advantages. There is therefore the opportunity for industry to use
the knowledge and expertise of academics; in turn, academics would gain access to real world
examples where new areas of research can be explored and tested.

One challenge to bridging this gap is that incentives in the two communities are often different.
Companies for example need to protect their IP in order to gain a competitive edge. Sharing ideas
and forming collaborations with the community can therefore be difficult. However, provided these
issues are recognised and discussed early when forming new collaborations, they can be managed.

4.5. Cross-Sector Knowledge Transfer

There were 34 delegates present at the data science workshop, with a variety of perspectives from
across a range of sectors, all with one thing in common – they want to get more out of data, and
they know that the mathematical sciences have the power to help. A community is forming.
Domains represented at the workshop included digital forensics, social media, environment,
engineering, technology, security, defence and aerospace.

There is an opportunity to take methods already in practice for one sector and test their suitability
for application in another. Sharing ideas across the domains, from cars to health to transport for
instance, introduces potential for increased growth, as opposed to just incrementally changing
current practice.

5. Challenges
There are many challenges and issues facing the data science community. This section highlights the
present challenges which caused concern to the delegates of the workshop and areas that could be
improved. Priority challenges, issues and areas to be addressed are:

 ensuring data is analysed such that the output gives actionable, priority options which can
form the basis for robust decisions;
 maintaining security and privacy of data whilst not discouraging collaborations and open
sharing of ideas and methods;
 making big data analysis compatible with computational science to leverage advances in fast
computation, data storage and distribution of data;
 ensuring data analysis and visualisation are dynamic, to encompass the continual stream of
data and the real-time inference that comes with it;

7
Data Science: Exploring the Mathematical Foundations

 creating a data science community which has access to high quality data, case studies and
expertise from various sectors across industry and academia;
 finding and hiring staff with the relevant expertise in data analysis as well as sector specific
knowledge.

These challenges are described in more detail below.

5.1. Big Data Machine Learning

A prominent challenge facing big data is the need for
Machine learning encompasses a valuable
more real-time streaming methods. Many data-driven set of techniques and methods that can be
sectors such as customer services have evolved to be used to evaluate data. Its methods focus
24/7 and therefore show a significant shift towards on the ability to learn without being
explicitly programmed and can therefore
requiring real-time data analysis. Big data
evolve and change when exposed to new
methodologies are required which can be tailored to data.
take advantage of timestamps and the data stream.
Machine learning is used in many
Research has been undertaken into streaming
applications to process large amounts of
methods but these ideas have not yet gone beyond data. It is therefore popular in the data
academic circles. Part of the challenge here is to science field and its range of applications
continues to grow. For example, machine
communicate these mathematical advances to the
learning has had much success for
wider community. predictive text messaging, Facebook photo
tagging and Google filtering.
Another challenge facing the mathematical sciences
within big data is the need to keep in line with Caution should however be applied when
relying on machine learning methods to
evolving technological advances. New hardware and inform significant decisions which can have
software platforms will continue to develop to deal high impact (positive or negative) on
with the evolving data landscape and the growth in people’s lives, for example the decision of
whether to give someone a loan or whether
data volumes. Bespoke data analysis tools written for
a patient qualifies for a heart by-pass
a specific hardware configuration may not be operation.
compatible with future advances and will therefore
become redundant if this is not accounted for.

5.2. Communication
It was evident from the workshop that communication between industry and the mathematics
community can be improved. Firstly, there is an issue of problem identification, formulation and
dissemination. There is a need to understand what it is that organisations want out of their data and
how this can be described to the relevant skills base in order to analyse the data most effectively.
Similarly, there is a need to elicit the benefits extracted from the data and help industry interpret
the mathematical results.

Visualisation is a powerful tool because it can give a clear idea of the message within the data. It can
assist the interaction between industry and the mathematical sciences, aid the explanation of the
outputs from data analysis, and inform the decision making process. There is, however, a challenge
in understanding the most appropriate visualisation methods and graphical displays to use. Most
importantly, the visualisation needs to be dynamic such that the most recent insights of the data are
presented.

8
Data Science: Exploring the Mathematical Foundations

During the workshop, it was articulated that in some cases industrial members will not be able to
understand the mathematics behind a solution, and that they would rely on a bespoke “black box”
solution to be produced by a mathematical scientist. Academics can only produce these solutions if
they know the types of problems that industry face. Often this may only come from looking at their
data. Communication through data is therefore important but can cause data security issues as
discussed in section 5.5 below.

It was also recognised that data is often horizontally distributed in vertically structured companies.
This shows the importance of effective communication within companies and across differing
internal groups, not just between industry and academia. Data scientists with management
consultant skills and the understanding to overcome communication issues are thus invaluable.

5.3. Validation and Verification

The area of data science is constantly evolving and new methods or techniques are frequently
required to adapt to the changing conditions, it therefore becomes harder to validate and verify the
methods and results. Mathematicians and statisticians are expected to create sound, defendable,
and auditable conclusions from the data. This is difficult if there are no case studies, or examples to
compare ideas, or processes in place to judge the outcomes. It is important to build a knowledge
base which has case studies and examples, and provides testing information and evidence to justify
results.

In addition, there are many mathematical methods which have not been widely applied in the area
of data science but could have the potential to bring considerable advantage. These include:

- Neural computing - Tropical geometry

- Topological data analysis - Pattern theory
- Algebraic statistics

5.4. Skills Shortage

The demand for data scientists has risen dramatically in recent years and many companies are
finding it hard to hire people with the relevant skills. This is partly due to the creation of new roles
and skill sets in industry that did not exist before.

It is uncommon to find experienced individuals who have strong sector specific skills who can also
apply cutting-edge mathematical methods of data analysis. There is great value in attracting
mathematicians into industrial areas to gain sector-specific knowledge or introduce sector-specific
expertise to data science techniques.

The industrialists at the workshop recognised the importance of the mathematical sciences when
drawing information from data, and highlighted their frustration of not having (or knowing how to
have) access to mathematicians and statisticians. The discussion uncovered that the implementation
of tools was not a problem within industry; what they need is access to people who can provide
insight. It also highlighted internal challenges within organisations, such as the need to break down
silos between different teams.

For mathematicians and statisticians, there are challenges of how to take the mathematical sciences
to industry and what are the best mechanisms for commercialisation of their knowledge.

9
Data Science: Exploring the Mathematical Foundations

There appears to be great demand in industry for the provision of advice on what the appropriate
mathematical and statistical techniques are, and for mathematicians to engage with the domain
experts to solve problems together. New collaborations can avoid non-experts taking on specialist
work, and instead create and support an environment where each expert contributes from their own
specialism.

5.5. Data Security and Anonymity

Collecting data for analysis carries with it the risk that the data might get into the 'wrong hands'.
Often the data which contains the most valuable insights is the most sensitive. The importance of
privacy and security is therefore a concern. This is not an aspect of data analysis that gets much
attention, but its importance is growing and needs to be addressed.

For sectors where the nature of work is classified or commercially confidential, collaboration
becomes harder. Operating in this manner can be a challenge because it reduces the opportunities
for open discussions and limits the available channels for help.

Furthermore, when fusing data within or between different sources, anonymity can, perhaps
inadvertently, be compromised. The stitching together of datasets by matching common features
can, accidentally or not, unveil sensitive information which in turn creates a security risk. This risk
needs to be managed to ensure the protection of individual and organisational rights.

Bringing together datasets also raises the question of ownership. Where do the boundaries lie when
merging data?

For open data, it is important to consider the freedom of use of that data along with the unintended
consequence of misuse. Legal requirements, such as those enforced by the Data Protection Act, may
not be scalable to current or future data use scenarios. The challenge of law enforcement, who and
how to monitor and maintain data usage and what a security process might look like, should be
considered.

6. Strategic Priorities
This section identifies activities that can bring added value to industry through the use of
mathematics and statistics within data science. The following are the priority areas that are
recommended:

 KTN and Innovate UK to help coordinate leadership for this area;

 improve access to high quality data;
 incentivise collaborations between industry and academia;
 encourage open research where research results are published alongside the data on which
they are based.

These are described in more detail below.

6.1. Data Science Community

It is recommended that a network of people is established who have a range of skills and expertise,
covering multiple industrial sectors and across disciplines within academic institutes. The delegates
at the workshop make a very strong and varied foundation for this network. At the base of this
network could sit the KTN. It would be important to define who takes clear ownership of this area

10
Data Science: Exploring the Mathematical Foundations

within the KTN and to define and understand their role. The network can provide assistance in
connecting the required skills to the relevant problem and maintain links between academia and
industry. Over time, this network can grow and become a clear and obvious place to seek help and
discuss data science issues.

Links can be formed between research councils and centres of excellence to build upon the network
and further improve the variety of skills, knowledge and expertise available. The network can offer
relevant training, seek out the right skills, identify techniques and influence the direction for
mathematics and statistics in data science.

In parallel, it is recommended that a central repository for various data sets and case studies is
created. These datasets should include benchmarking and examples of good practice. Giving access
to data, examples and case studies will improve our understanding, support new research, and
create a better insight into how mathematics can bring value. Relevant open source repositories
already exist and should be used as a source of complementary information (whilst being careful of
differing terminologies between communities).

6.2. Conferences and Workshops

Conferences, workshops and meetings provide further networking opportunities, encourage
learning across different sectors and disseminate updates on applications of methods. These events
should be organised on a regular basis and advertised widely, attracting participants from multiple
disciplines and a variety of industrial sectors to inspire cross fertilisation of ideas. Individual events
would benefit from focusing on specific application areas or specific problem categories.
Opportunities for different communities to interact in an informal way can also be of benefit.

6.3. Education
Demand for data science skills is increasing. To cover the skills shortage, it is recommended that we
build capacity through education and encourage the relevant skills to be taught at University level.
This requires participation and commitment from academics to influence the courses that are set
across a variety of disciplines.

In the shorter term, reference material can be collated to teach and inform the community when
dealing with data. A catalogue of data types could be created which note the different mathematical
and statistical techniques that are relevant. This would be generated by mathematical scientists but
in a way that can act as a guide and reference for industry. Because of the extensive nature of data
science, it would be hard to cover all types of data and associated techniques but at least a
catalogue would be able to provide industry with a breadth of knowledge that would not normally
be accessible, with the opportunity to contact the relevant mathematical scientists who can help.

6.4. Encouraging Mathematical Activities

Open source competitions and challenges are recommended. Competitions that pose a data science
problem and require mathematical techniques to solve it (similar to the Kaggle competitions for
machine learning) can attract interest to the area, introduce new ideas and form relationships
between relevant sectors. Moreover different approaches can be directly compared.

Another option is to host a Study Group, where industrial problems are presented and a variety of
students and academics come together to discuss how they can be solved. Study Groups have been

11
Data Science: Exploring the Mathematical Foundations

very successful not only at helping to solve industrial problems with mathematics but also with
bringing groups together and bridging the gap between mathematics and industry.

Mathematical sciences should take advantage of current opportunities. For example there is an
opportunity to engage with Europe through Horizon 2020 and there will be opportunities with the
planned Alan Turing Institute to expand, research, educate and transfer knowledge in data science.

7. Recommendations
Section 6 recommends activities that can be jointly undertaken to increase added value for industry.
It is important when undertaking these activities that consideration is given to the nature of
engagement. Successful activities will share and manage expectations of timescales, intellectual
property, the academics' roles, industrial responsiveness to the problems and solutions, how the
data is to be distributed and stored, and how future engagements might develop.

With this in mind, we recommend that any initiatives undertaken to address the data science
challenge should be directed by the following key features:

 emphasise co-development of solutions between data users, data analysts and data
providers;
 develop mechanisms that identify and kick-start early stage opportunities, encouraging new
ideas to be tested quickly;
 encourage commitment of industrialists to exploit these opportunities through further
development, and document the impact;
 share experiences across domains, and promote good practice in data cleanliness and data
sharing;
 operate in a way that does not get tied up in intellectual property or legal requirements
before any progress can be made.

If these principles are followed, the mathematical sciences will be in a much stronger position to
exploit the opportunities within data science and bring added value to industry.

8. Acknowledgements
This report was written by the Smith Institute, with many thanks to the delegates (Appendix B) for
their attendance, input and discussions at the “The Application of the Mathematical Sciences to the
Underpinning Foundations of Data Science” Workshop on the 23rd July 2014.

9. References
[1] Durrett, R. (2010) Probability: Theory and Examples. Fourth Edition. Cambridge University Press

[2] Rabiner, L.R. (1989) A tutorial on Hidden Markov Models and Selected Applications in Speech
Recognition. Proceedings of the IEEE, 77, 2, 257-286

[3] Grindrod, P., Parsons, M.C., Higham, D.J., Estrada, E. (2011) Communicability across evolving
networks. Physical Review E, 83

[4] Gomez-Gardenes, J., Reinares, I., Arenas, A. & Floria, L.M. (2012) Evolution of Cooperation in
Multiplex Networks. Nature Scientific Reports, 2, 620

12
Data Science: Exploring the Mathematical Foundations

[5] Arel, I., Rose, D.C. & Karnowski, T.P. (2010) Deep Machine Learning – A new Frontier in Artificial
Intelligence Research. IEEE Computational Intelligence Magazine

[6] Ghosh-Dastidar, S. & Adeli, H. (2009) Spiking neural networks. International Journal of Neural
Systems, 19, 04, 295

[7] Hinton, G.E. (2009) Deep belief networks. Scholarpedia. 4,5,5947

[8] Schoot, R., Kaplan, D., Denissen, J., Asendorpf, J.B., Neyer, F.J. & Aken, M. (2013) A Gentle
Introduction to Bayesian Analysis: Applications to Developmental Research. Child Development, 1-19

[9] Michie, D., Spiegelhalter, D.J. & Taylor, C.C. (1994) Machine Learning, Neural and Statistical
Classification. Cambridge

[10] Finley, T. & Joachims, T. (2005) Supervised Clustering with Support Vector Machines.
Proceedings of the 22nd International Conference on Machine Learning, Germany.

[11] Hand, D., Mannila, H. & Smyth, P. (2001) Principles of Data Mining. Massachusetts Institute of
Technology

[12] Koller, D. & Friedman, N. (2009) Probabilistic Graphical Models: Principles and Techniques. MIT
Press

[13] Lum, P.Y., Singh, G., Lehman, A., Ishkanov, T., Vejdemo-Johansson, M., Alagappan, M. Carlsson,
J. & Carlsson, G. (2013) Extracting insights from the shape of complex data using typology. Nature
Scientific Reports, 3, 1236

[14] Mikhalkin, G. (2006) Tropical geometry and its applications. Proceedings of the International
Congress of Mathematicians. http://arxiv.org/pdf/math/0601041v2.pdf [21 November 2014]

[15] Maclagan, D. & Sturmfels, B. (2013) Introduction to Tropical Geometry. Citeseer.

http://homepages.warwick.ac.uk/staff/D.Maclagan/papers/TropicalBook23.8.13.pdf [21 November
2014]

[16] Thelen, E. & Smith, L.B. (2005) Dynamic Systems Theories.

http://www.iub.edu/~cogdev/labwork/handbook.pdf [21 November 2014]

[17] Smola, A. & Vishwanathan, S.V.N (2008) Introduction to Machine Learning. Cambridge
University Press

[18] Schatz, M., Low, T., Van de Geijn, R. & Kolda, T. (2014) Exploiting symmetry in tensors for high
performance. SIAM Journal of Scientific Computing

[19] Byrd, R. H., Hansen, S.L., Nocedal, J. & Singer, Y. (2014) A Stochastic Quasi-Newton Method for
Large-Scale Optimization, arXiv. http://arxiv.org/pdf/1401.7020v1.pdf [21 November 2014]

[20] Gleich, D. F., (2014) PageRank beyond the web, Preprint on arXiv.
http://arxiv.org/abs/1407.5107 [21 November 2014]

13
Data Science: Exploring the Mathematical Foundations

Appendix A - Workshop Agenda

10:00 10:30 Registration KTN

10:30 10:45 Welcome KTN, SI
Session 1: Setting the Scene
10:45 11:15 Big Data in ICT Ian Osborne, KTN
11:15 11:45 Data Analytics at GCHQ Bill Oxbury, GCHQ
11:45 12:15 Customer Facing Industries Peter Grindrod, Oxford
12:15 12:45 Social Media Data Peter Laflin, Bloom Agency
12:45 13:45 Lunch and Networking
13:45 14:45 Session 2: Challenges and Perspectives Delegates
14:45 15:15 Coffee and Network
15:15 16:15 Session 3: Solutions and Strategies Delegates
16:15 16:30 Conclusions and Final Points KTN
16:30 Coffee and Close

14
Data Science: Exploring the Mathematical Foundations

Appendix B – Delegate List

First Name Surname Affiliation

Ali Anjomshoaa Forensic Science SIG
Andrew Barnes Home Office
Christoph Best Google
Philip Bond Council for Science and Technology
Matt Butchers The Knowledge Transfer Network
David Calder The Knowledge Transfer Network
Felicity Carlyle The Knowledge Transfer Network
Simon Cooper Welsh Government
Mike Dewar NAG
Mark Doutre Lockheed Martin
Robert Downes CASA – University College London
Torran Elson Smith Institute
Hannah Fry University College London
Rob Gill Deutsche Bank
Peter Grindrod University of Oxford
David Hand Imperial College London
Des Higham University of Strathclyde
Jonathan Hogg Science & Technology Facilities Council
Dan Jabry CrowdEmotion
Zoe Kelson Smith Institute
Peter Laflin Bloom Agency
Michelle Ledbetter Smith Institute
Tom McCutcheon MOD/Dstl
Jonathan Mitchener Technology Strategy Board
Peter Murray The Knowledge Transfer Network
Ian Osbourne The Knowledge Transfer Network
Bill Oxbury GCHQ
Tim Palmer University of Oxford
Mark Pitman Medical Research Council
Sanjiv Sharma Airbus
Andrea Sharpe Natural Environment Research Council
Patrick Wolfe University College London
Jeremy Yates University College London

15