Statistical methodology and theory for the Big Data era (Ext.)

This is an extension of the Fellowship 'New challenges in
high-dimensional statistical inference' (EP/J017213/1).

The project will involve developing methodology and theory for a range
of modern statistical problems from high-dimensional and nonparametric
statistical inference. Such problems lie at the heart of modern
data science, an area where the UK urgently needs to grow its capacity.

Specific challenges to be addressed include the development of methods
for dimension reduction, variable selection and uncertainty
quantification, among several others. These are some of the
fundamental problems faced by practitioners in the Big Data era, and
demand highly innovative solutions. Our proposed techniques will be
proven to be robust via appropriate theoretical justification, and
will be implemented in free, open source software so as to maximise
the impact of the research.

Big Data are transforming the way we live. These days, we can access
an almost unimaginably large wealth of knowledge by typing a few
keywords into an internet search engine and use apps on our mobile
telephones or other smart devices to monitor or improve our health.
Many recent advances in healthcare are partly due to improved, highly
data-intensive scanning equipment in hospitals, and the development of
new, effective drug treatments, which have been the result of
extensive scientific study with data at its core.

This proposal addresses some of the fundamental and important
statistical challenges that arise in handling the modern data sets
that routinely arise in the applications above, as well as many
others. These challenges include dimension reduction, variable
selection and uncertainty quantification. It therefore has the
potential for high societal and economic impact, both in the immediate
applications considered and through later transfer of the innovative
new methods that will be developed.

In the UK, we are currently facing a great shortage of well-trained
data scientists, across both science and industry. This proposal will
go some way towards addressing this deficiency, since the two
post-doctoral research associates employed will be exposed to
cutting-edge problems in the field, and will acquire the crucial
skills that are in so much demand in many sectors of the economy,
including academia, and the technology and pharmaceuticals sectors.

I am the main organiser of a six-month Isaac Newton Institute
programme on 'Statistical Scalability' that will form one of the major
impact activities of the proposal. End users of modern statistical
methodology and theory from science and industry will be integrated
into the programme (for instance, through an `Open for Business' day
that will be co-organised with the Turing Gateway for Mathematics) to
ensure wide dissemination and cross-fertilisation of the ideas. They
will therefore benefit from exposure to the ideas of the top
researchers who will participate in the programme, and can translate
the key messages into their own communities.

Through the collaboration with Martin Bogsted and his cancer genetics
group, we will improve understanding of the prognostic value of gene
expressions from various B-cell malignancies. In particular, we will
ascertain which genes are associated with short survival for specific
diseases and which have a common association with short survival for
multiple diseases. The societal benefit will be shortcuts in
identifying the applicability of drugs, which have the potential to be
incorporated into clinical trials.