Research Fellows

Employed by our Joint Venture Universities, the inaugural Alan Turing Institute Research Fellows will pursue research based at the Institute hub for the next 3-5 years.

Profiles for each Research Fellow are below.

Joshua Loftus

Cambridge

BIO

Joshua Loftus is a statistician and data scientist with research interests in high-dimensional statistics and post-selection inference, and more broadly in statistical theory and methodology driven by applications in biomedicine or motivated by social good. He earned his PhD at Stanford University and plans to join the statistics faculty at NYU Stern in the fall of 2017.

 

RESEARCH

Post-selection inference is a new area of statistics concerned with adjusting inferences to account for model selection. Joshua’s research focuses on complex model selection procedures like cross-validation, and the development of software implementations that conduct inferences such as significance testing for the variables included in the final model. Of particular interest is the application of these methods in causal inference to search for heterogeneous treatment effects and report valid inferences for those effects without requiring inefficient sample splitting.

Brooks Paige

Oxford

BIO

Brooks Paige moved to the UK in 2013 to undertake a D.Phil. in Engineering Science at the University of Oxford. He previously lived in New York City, where he completed an M.A. in Statistics at Columbia University and worked professionally for several years as a software developer and web designer. In the distant past, he was an undergraduate at Amherst College, and remains a strong proponent of a liberal arts science education. http://www.robots.ox.ac.uk/~brooks/

 

RESEARCH

The emerging field of probabilistic programming aims to reduce the technical and cognitive overhead for writing and designing novel probabilistic models, by introducing a specialized programming language as an abstraction barrier between modeling and inference. Brooks’ research focus is on developing general-purpose Bayesian inference algorithms which can be applied automatically to generative models written as probabilistic programs, with a particular emphasis on sequential Monte Carlo methods and scalable variational inference.

Dong Nguyen

Edinburgh

BIO

Dong Nguyen has a master’s degree from Carnegie Mellon University and is currently finishing her PhD at the University of Twente. She has interned at Google, Microsoft Research and Facebook. She works on text mining methods that can help answer questions from the social sciences and the humanities. Her research was featured in the New York Times and Time Magazine. http://www.dongnguyen.nl/

 

RESEARCH

The rise of social media brings exciting opportunities to study social phenomena through large-scale text analysis. So far, the dominant approach for text analysis of social media data is based on the concept of language as a means to convey information. However, people also use language to construct their identities, and to build and maintain social relationships. In her research, he focusses on using computational models to study the social dynamics of language use in social media.

Hemant Tyagi

Edinburgh

BIO

Hemant Tyagi completed his PhD at the Institute of Theoretical Computer Science, ETH Zurich, in 2016. Earlier, he obtained a Masters degree in Communication Systems at EPFL in 2011, and a Bachelors degree in Electrical & Electronics Engineering at NIT Surathkal, in 2006.

 

RESEARCH

Hemant’s research interests broadly revolve around the areas of: low dimensional models for high dimensional data, compressive sensing, approximation theory and high dimensional statistics. He is also interested in (non-) convex optimization as well as online optimization (multi armed bandits etc.).

Chris Russell

Edinburgh

BIO

Chris Russell is a research fellow at the Alan Turing Institute and associated with the University of Edinburgh.  Following a PhD in computer vision with Prof. Philip Torr, he worked with Prof. Lourdes Agapito at UCL. His primary research focus is the application of novel optimisation techniques to problems of machine learning and computer vision, particularly 3D reconstruction from video.

 

RESEARCH

The aim of his fellowship is to make algorithms that see with structure, i.e. reliable algorithms that reason in 3D about a dynamic and interactive world from video footage, and generate vivid and lifelike 3D models from video. The principle idea guiding his research is the notion that the world is a dynamic 3D place, and if we want computers to reason usefully about the world, they should do so in 3D. Building accurate 3D models of moving and deforming objects from video will have immediate impact for driverless cars, robotic interactions, and in the special effects and movie industry.

Armin Eftekhari

Edinburgh

BIO

Armin Eftekhari received his PhD from Colorado School of Mines in 2015, under the supervision of Michael Wakin. Before joining the Alan Turing Institute, he was an ICES postdoctoral fellow at the University of Texas at Austin.

 

RESEARCH

The late statistician George Box once wrote that “all models are wrong, but some are useful.” Indeed, to make sense of data, we must necessarily impose some structure upon it. For example, we often assume that digital data is collected from bandlimited sources. Or, in compressive sensing, a small number of measurements suffice to reconstruct a sparse signal. Making inferences from (often noisy, incomplete, and high-dimensional) data by imposing appropriate structures has been the main theme of Armin’s research, and he hopes to continue this line of work at the Alan Turing Institute, with more emphasis on practical aspects.

https://sites.google.com/site/armineftekhari

Mihai Cucuringu

Oxford

BIO

Mihai received his PhD in Applied and Computational Mathematics at Princeton University in 2012, supervised by Amit Singer. During 2013-2016 he was a CAM Assistant Adjunct Professor in Computational Applied Mathematics at UCLA. His research concerns the development and mathematical analysis of algorithms for large networks and data analysis, with applications to various problems in engineering, machine learning, and biology.

 

RESEARCH

Mihai will investigate spectral algorithms for analysis of signed (multilayer) networks, with applications to correlation networks. Building on the Vector Diffusion Maps framework, he will consider matrix and vector-based affinities that capture different types of relationships between nodes of a graph, thus moving beyond the traditional scalar affinity. Such approaches are useful for obtaining low-dimensional representations of networks arising from high-dimensional time series data. Additional strands concern variations of the group synchronization problem, and ranking from noisy pairwise comparisons. The above problems share an important feature: they can all be solved by exploiting the spectrum of their corresponding graph Laplacian.

Efthymia Tsamoura

Oxford

BIO

Efthymia Tsamoura received her BSc in 2007 and her PhD in 2013 with honours both from the Computer Science Department of Aristotle University of Thessaloniki, Greece. Since June 2013 she is a postdoctoral researcher in the Computer Science Department of University of Oxford, working on problems related to querying semantically interrelated resources. Her research interests lie in the fields of data integration, query answering under constraints, distributed query optimization, multi-objective query optimization and grid computing. Efthymia received several distinctions and awards both during her undergraduate and postgraduate studies.

 

RESEARCH

Today a vast amount of bioinformatics data resides on the web. Yet the usefulness of these resources is limited by the difficulty in harnessing the data to answer user information needs. In collaboration with researchers from Turing, the University of Oxford, and the European Bioinformatics Institute (EBI), she will create tools and techniques enabling biologists to more easily integrate web-based data into their activities. This will involve building systems that allow scientists to use the plethora of web-based data without expertise in the resources that want to access. While making the use of data simpler, the tools that she will create will also make them run more efficiently, by giving scientists access to the power of large-scale infrastructure.

Vidit Nanda

Oxford

BIO

Dr Vidit Nanda is an applied and computational algebraic topologist. Before starting at the Alan Turing Institute, he worked a post-doctoral research fellow at the University of Pennsylvania and as a PhD candidate in mathematics at Rutgers University. http://www.sas.upenn.edu/~vnanda/

 

RESEARCH

Dr Nanda develops algebraic-topological theories, algorithms and software for the analysis of non-linear data and complex systems arising in various scientific contexts. In particular, he employs discrete Morse-theoretic techniques to substantially compress cell complexes built around the input data without modifying core topological properties. His recent work has involved excursions into computational geometry, cellular sheaf theory and higher-categorical localization.

Matt Kusner

Warwick

BIO

Matt J. Kusner was a visiting researcher at Cornell University under the supervision of Kilian Q. Weinberger and received his Ph.D. in Machine Learning from Washington University in St. Louis. His work is in the areas of privacy, budgeted learning, model compression, and Bayesian optimization. He is from Iowa City, Iowa, USA and is married to the wonderful Sonia Rego.

 

RESEARCH

Matt’s research aims to address the disconnect between state-of-the-art machine learning models, and models that are often used to solve real-world problems. Frequently, in real-world settings the modeler is confronted with a trade-off between maximizing an objective (e.g., returning more accurate results) and minimizing a budget (e.g., producing predictions in under a millisecond). His research considers three specific types of budgets: time, space, and privacy. These feature in recommendation, face recognition, bankruptcy prediction, stock market modeling, and real-time machine translation. Directly addressing these real-world trade-offs at an optimization level results in algorithms that are simultaneously practical and accurate.

Nathanaël Fijalkow

Warwick

BIO

Nathanaël Fijalkow obtained his PhD in 2015 in Computer Science, entitled “Counting and Randomising in Automata Theory”, jointly awarded by Paris 7 and Warsaw University under the supervision of Thomas Colcombet and Mikołaj Bojańczyk. He joined the University of Oxford in 2016 as a Research Assistant in the Verification group. His research interests are in Logics, Automata Theory, Games, Stochastic Systems, Formal Methods, Verification and Complexity.

 

RESEARCH

Nathanaëls Research Fellowship at the Alan Turing Institute will allow him occasion to broaden his perspectives; he plan to work on Streaming Algorithms, Robustness for Stochastic Systems, Probabilistic Databases and Machine Learning. His ambition is to provide theoretical foundations for these questions, and to demonstrate their values through practical applications.

Catalina Vallejos

UCL

BIO

Catalina Vallejos is a Bayesian statistician. She graduated from the University of Warwick (PhD in Statistics, 2014) and from the Pontificia Universidad Catolica de Chile (BSc and MSc in Statistics, 2010). Before joining the Turing, she worked as a postdoctoral researcher on statistical genomics at the MRC Biostatistics Unit and the EMBL European Bioinformatics Institute.

 

RESEARCH

Catalina’s research at the Turing will focus on applied Bayesian statistical methodology, with emphasis on the development of open-source analysis tools for high-dimensional datasets with complex structure. Within a Bayesian hierarchical modelling framework, she will develop integrative methods to combine information from different data sources and data types, to disentangle multiple interacting sources of variability. Her work will serve as a bridge between advanced statistical methods and applied areas of research, including strong collaborations with biologists and other scientists.

Tomas Petricek (Visiting Researcher)

Cambridge

BIO

Tomas is a Visiting Researcher at the Alan Turing institute, working on data-driven storytelling. He is building tools that integrate with modern data sources (open government data, data published by citizen initiatives) and let users easily create analyses and visualizations that are linked to the original data source, making the analyses more transparent, reproducible, but also easy to adapt. His early work on the project can be found at http://thegamma.net.

RESEARCH

Tomas’ many other interests include open-source and functional programming (he is an active contributor to the F# ecosystem), programming language theory (his PhD thesis on “coeffects” develops a theory of context-aware programming language language), but also understanding programming through the perspective of philosophy of science.

Yarin Gal (Visiting Researcher)

Cambridge

BIO

Yarin is a research fellow at St Catherine’s College, Cambridge, and part-time fellow at the Alan Turing Institute. He obtained his PhD from the Cambridge machine learning group, working with Zoubin Ghahramani and funded by the Google Europe Doctoral Fellowship. Prior to that he studied at Oxford Computer Science for a Master’s degree under the supervision of Phil Blunsom.

RESEARCH

Yarin’s interests lie in the fields of linguistics, applied maths, and computer science. Most of his work is motivated by problems found at the intersections of these fields. In his current research he develops Bayesian techniques for deep learning, with applications to reinforcement learning. In the past he has worked on Bayesian modelling, approximate inference, and natural language processing. This work included Bayesian nonparametrics, Gaussian processes, inference algorithms for big data, and work on machine translation.