Research
Mobility data science
funded by Toyota Mobility Foundation, £350k in 2020.
There is no Excel for mobility data, no common platform for
testing ideas and sharing them. The issue is that mobiility data is
oriented around user histories, whereas Excel is oriented around tabular data. I am building a new
class of tool which will enable end users to make sense of mobility data, especially new services
such as rideshare and dockless bikes. I am also developing new theory for modelling such services:
theory that combines models of user choice with models of system dynamics.
Probabilistic machine learning.
I am interested in how neural networks can be used as building
blocks for probabilistic modelling, building on the way that
generations of statisticians have used linear models.
In particular, I have developed a new way to train neural networks, to enable them
to express how confident they are in their predictions.
Explainability and the philosophy of machine learning.
Over the past year I have collaborated with a lawyer, and
uncovered some fascinating parallels between machine learning and
legal philosophy.
Employment
Computer Laboratory, Cambridge, 2017–: university
lecturer
, working in the field of data science.
Alan Turing Institute, 2018–: fellow,
working on data science to model shared vehicle fleets and the future of transport,
as part of the Urban Analytics research theme.
Urban Engines (a startup based in Los Altos, CA), 2011–2016: chief data scientist.
We built a platform for visualising and analysing big data
about things that move: commuters, trains, buses, taxis, delivery
fleets, etc. Customers included several major cities and transport
infrastructure providers. Technologies used include
Spark, d3, Postgres, and Amazon's
AWS. My work involved all levels of the stack: talking with
customers, designing system capabilities, data architecture, devising
visualisations and inference algorithms, and detailed coding in R,
Python, Scala, Javascript. The
company had 22 employees, 10 of them in the data science group.
Urban Engines was acquired by Google in 2016.
Electrical Engineering, Stanford: visiting professor in 2011,
consulting professor in 2016.
In 2011 I worked with Prof. Balaji Prabhakar on
societal networks—networks that combine real-world
infrastructure and people. I worked on health
incentives for Accenture employees, and transit incentives for
Singaporean commuters. (This work led into Urban Engines, which
Prof. Prabhakar cofounded.)
In 2016, Prof. Prabhakar
and I are co-teaching a course on Big Data for Things that Move, to
graduate students in computing and engineering.
UCL, London, 2004–2011: Royal Society university research
fellow,
based in the Networks and Systems group in the Computer Science
department.
I came to UCL with the goal of translating theoretical work on
congestion control, begun by Prof. Frank Kelly in Cambridge,
into a practical system. With Prof. Mark Handley
and others in the EU-funded Trilogy project,
we created MPTCP (Multipath TCP), which became an IETF standard
and is used by Siri on the iPhone.
Statistical consulting, 2006–
for TauRx Therapeutics, a
biotech startup based in Aberdeen, working on Alzheimer's Disease.
This has involved interacting with medics,
scientists, clinical research organisations, regulators, valuation
consultants, and investment bankers, and has spanned Phase 2
and Phase 3 clinical trials. I have analysed data and
advised on clinical trials, psychometrics, animal experiments, business development,
risk and valuation.
This work has given me a deep appreciation of statistics as a
form of rhetoric, not just a tool for mathematical modelling.
Trinity College, Cambridge, 1999-2004: Junior Research Fellow,
an independent research position. I was based in the
Statistical Laboratory in
the University of Cambridge. I worked on probability theory for
queueing networks, with application to Internet switches.
Selected outputs
T.D.Grant and D.Wischik (2020).
The path to AI: law's prophecies and the conceptual foundations
of the machine learning
age.
Palgrave.
The big novelty of machine learning—the elevation of
prediction over explanation—was prefigured by Oliver
Wendell Holmes Jr, a judge who reshaped American legal thinking at
the end of the 19th century. The ensuing jurisprudential debates
have much to offer machine learning.
T.D.Grant and D.Wischik (2020).
Show us the data: privacy, explainability, and why the law can't
have both. To appear in George Washington University Law Review.
The GDPR requires that explanations be given of decisions given by
machines—but for machine learning algorithms the only
meaningful explanation involves sharing the training data.
D.Wischik (2018).
The price of choice:
models, paradoxes, and inference for ‘mobility as a service’.
Allerton.
This paper unifies discrete choice modelling with network flow optimization, via
information theory.
It points the way to a comprehensive economic understanding of Mobility as a Service.
N.Gomes, D.Merugu et al. (2012).
Steptacular: an incentive mechanism for promoting wellness.
COMSNETS NetHealth. This is the only publication to come out of my
work on incentives—but the system I built is running at
www.travelsmartrewards.com, has 330,000 users, and has paid out $10M
Singapore dollars over four years.
D.Wischik, C.Raiciu, and A.Greenhalgh (2011).
Design,
implementation and evaluation of congestion control
for multipath TCP. NSDI, winner of best paper award.
This work has been standardized as an Internet Experimental
Standard,
RFC 6356.
D.Shah and D.Wischik. Switched networks with maximum weight policies:
fluid approximation and multiplicative state space
collapse. Annals of Applied Probability (2012).
Fluid models of congestion collapse in overloaded switched
networks. Queueing Systems (2011).
C.M. Wischik, D.J. Wischik, J.M.D. Storey, C.R. Harrington (2010). Rationale
for tau aggregation inhibitor therapy in Alzheimer's disease and other
tauopathies. Chapter in Emerging drugs and targets for Alzheimer's
disease, vol. 1, ed. A. Martinez, RSC Drug Discovery Series.
I am a co-author on three patents relating to this work.
G.Raina, D.J.Wischik (2005).
Buffer
sizes for large multiplexers: TCP queueing theory and instability
analysis. This work lead to a DARPA grant, and to a series of
letters in ACM Computer Communication Review, co-authored with
Nick McKeown and Don Towsley.
A.Ganesh, N.O'Connell, D.J.Wischik (2004).
Big Queues,
a book. Awarded the 2004 Best Publication Award by the Applied
Probability Society of INFORMS.