Share on Facebook Tweet on Twitter Share on LinkedIn Share by email
2009 PhD Scholars

Abhijit Karnik
University of Bristol, United Kingdom

Supervisor: Dr Sriram Subramanian
Microsoft Research supervisor: Alex Butler, Shahram Izadi

Research title: Using optical devices to enhance user interactions on an interactive surface

Research summary: Today’s touch-screen technology potentially allows multiple users to simultaneously interact with one another and with digital content by using their whole hand to engage in the interaction but it’s limited in the sense that all users have to share the same visible content. However, most of today’s touch-based systems only support single views for their interaction. In other words, these systems do not allow multiple users to view information that is customized to their view on the same interactive surfaces. There is limited systematic study of combining multi-touch with multi-visibility. There have been a few one-off point-designs (proof of concept systems to show that it’s possible to build such systems) but no systematic investigation into the benefits and limitations of combining multi-touch with multi-visibility has been performed. Here we propose to systematically investigate the design of interactive surfaces that use optical devices (like lenticular lenses and polarizers) to support multi-touch and multi-visibility.

Adam Gundry
University of Strathclyde, United Kingdom

Supervisor: Dr Conor Thomas McBride
Microsoft Research supervisor: Simon Peyton Jones

Research title: Haskell Types with Numeric Constraints

Research summary: This PhD project seeks to investigate the practical and theoretical impact of extending Haskell’s type system with numeric expressions (representing sizes, ranges, or costs, for example) and constraints capturing richer safety properties than are currently managed by static typing. There are three strands to the project: (1) to investigate type inference with numeric constraints, (2) to investigate new programming structures, patterns, and techniques which exploit numeric indexing, and (3) to study the performance benefits derivable from richer guarantees. There are considerable opportunities for a bright student to bring significant benefits to developers using Haskell, a language with increasing industrial traction – not least at Microsoft, where its flagship compiler is maintained, and where it plays a key role in a variety of cutting-edge projects. Moreover, Haskell is an established staging post for technology on its way to deployment in mainstream languages, e.g. C#.

Alice Boit
University Potsdam, Germany

Supervisor: Prof. Dr. Ursula Gaedke,
Microsoft Research supervisor: Lucas Joppa

Research title: Visualising the mechanisms influencing ecosystem stability in quantitative food webs

Research summary: The stability of ecosystems has been a central research topic in ecology since the 1950’s and has been found to depend on a multitude of biotic and abiotic factors. The proposed PhD project will continue this research by helping to answer, ‘How do properties of populations influence ecosystem stability?’ In particular, this project aims to account for differences in the importance of individual feeding interactions and to assess the influence of prey edibility and predator food selectivity on ecosystem stability from a population and community level perspective. Combining theory and empirical data to simulate and visualize food webs, current software tools will be extended with new functionality to clarify and visually communicate how food web structure is influenced by its dynamics and vice versa. Thus, new simulations and three-dimensional food web visualisations developed in the project will help scientists better explore patterns of structural and dynamical properties and help predict ecosystem dynamics and stability.

Andrea Flack
University of Oxford, United Kingdom

Supervisor: Dr Dora Biro
Microsoft Research supervisor: Robin Freeman

Research title: Collective decision-making in avian navigation

Research summary: Animals that live in social groups must make joint decisions about many aspects of their daily lives, and the mechanisms that mediate such collective decision-making have generated a great deal of theoretical interest in recent years. However, empirical evidence has been almost entirely lacking, and the extent to which mathematical models of group decision-making, conflict resolution, and social learning translate into the real world has remained unresolved. The proposed project combines an experimental approach—the tracking of homing pigeons using miniature GPS technology—with a mathematical and computational modelling framework to examine the mechanisms and consequences of collective motion in co-navigating birds. Homing pigeons provide a unique and ideal study system where individuals’ possession of information can be controlled, social interactions regulated, and the flow of information quantified. A variety of modelling techniques will be tested for validity in explaining the empirical findings, and the generalities for collective decision-making in a wide variety of contexts will be considered, with the ultimate aim of expanding our knowledge of how collectives are shaped by the individuals of which they are composed.

Antje Beyer
University of Cambridge, United Kingdom

Supervisor: Dr Gos Micklem
Microsoft Research supervisor: Jasmin Fisher

Research title: Computational insights into the effect of genetic variations on C. elegans vulval development

Research summary: Computational modelling of biological systems is becoming increasingly important in efforts to better understand complex biological behaviours. ‘Executable Biology’ is a pioneering approach focused on the design of executable computer programs that mimic biological phenomena. We have previously established an executable model of C. elegans vulval development that includes key components of the RAS/MAPK and LIN-12/Notch pathways as well as the crosstalk between the two pathways, which is essential for the process of vulval precursor cell fate determination. The aim of this PhD proposal is first to extend the current model to include key components of the Wnt signalling pathway that act in parallel to RAS/MAPK signalling as well as transcription factors (e.g., sur-2, lin-39, lin-1 and lin-31) that act downstream to the MAPK pathway. Once this model has been built, genomic, proteomic and cellular data obtained by the Hengartner, Poulin, and Jansen labs will be integrated into the revised model. Since the genomic analysis will also generate quantitative data in the form of expression levels or phenotype penetrance (obtained by the Hajnal and Hengartner labs) we also aim to develop a hybrid model incorporating some quantitative data such as expression levels for selected key components in the studied signalling pathways. In addition, as many of the known regulators of vulva development also control RAS/MAPK signalling during germ cell development, we intend to expand our vulva development model to include germ cell development.

Antonia Masucci
Supélec, France

Supervisor: Prof. Merouane Debbah
Microsoft Research supervisor: Peter Key, Bozidar Radunovic

Research title: Advanced mathematical tools for the design of cognitive radios

Research summary: It has already become a common understanding that current mobile communication systems do not make full use of the available spectrum, either due to sparse user access or to the system’s inherent deficiencies, as shown by recent reports from the Federal Communications Commission (FCC). It is envisioned that future systems will be able to opportunistically exploit those spectrum ‘left-overs’, by means of knowledge of the environment and cognition capability, in order to adapt their radio parameters accordingly. Such a technology has been proposed by Joseph Mitola in 2000 and is called cognitive radio. Due to the fact that recent advances on micro-electronics and computer systems are pointing to a -not so far- era when such radios will be feasible, it is of utmost importance to develop adequate mathematical tools for designing and optimizing cognitive radios which can extract and understand the wireless network on their own. The goal of this PhD is to propose new mathematical tools based on free de-convolution and random matrix theory techniques (and for which the PhD advisor has made some important contributions in many other areas) to extract information from the network

Daniel Gallardo
Universitat Pompeu Fabra, Spain

Supervisor: Dr. Sergi Jordà
Microsoft Research supervisor: Shahram Izadi

Research title: Exploring expanded control possibilities within SecondLight

Research summary: In recent years we have seen a proliferation of tangible and tabletop interfaces, several of them already becoming real products. And yet, much before the implications and the potential of these new types of interfaces get fully explored and exploited, much less well understood, newer technologies keep bringing amazing new-fangled possibilities. This is the case of SecondLight, a technology developed by Microsoft that combines all the potentiality of surface computing using tangibles and extends it “beyond the display”, allowing to interact also from a distance. We propose to investigate the new interaction possibilities this outstanding and almost magical technology may bring. Following the work developed by our research team within the reactable project, all this research will be oriented towards the idea of expanded control sharing. That is, instead of focusing on a data-centric view of interaction, we will mostly consider representational forms as resources for action. Instead of relying on the transmission and sharing of data, we will be looking for solutions that emphasise user control, creativity, and social action with interactive tools.

Frédéric Besse
Supervisor: Dr Jan Kautz, University College London

Microsoft Research supervisor: Andrew Blake
Research title: Automatic enhancement of digital photographs

Research summary: The main objective of this proposal is the automatic enhancement of digital photographs. In particular, we want to improve a snapshot’s “look” by learning and imposing statistics from good photographs, as well as enhance the snapshot by cropping it according to basic composition rules that are also inferred from good photographs (and their salient features). The main challenges are the choice of statistics and how they can be enforced or applied. Our hypothesis is that data-driven, automatic enhancement of snapshots is viable and we will validate the results with user studies.

Ioannis Psorakis
University of Oxford, United Kingdom

Supervisor: Prof. Ben Sheldon
Microsoft Research supervisor: Robin Freeman

Research title: The form and function of dynamic sociality in a wild bird population

Research summary: This proposal aims to use an exceptional data set from a large number of marked individual birds, tracked automatically over several winters, to explore the formation and dynamics of social groups of foraging individuals. The aim will be, first, to derive computationally intensive methods for identifying groups; second to understand the stability of those groups, and the way that the characteristics of individuals making up the groups influence their behaviour and composition, and third, to test hypotheses relating to the fitness consequences of sociality. The study will be embedded in the context of a long-term population study which provides richly detailed information about individual life-histories, overlain on environmental and genetic data.

Isabel Rosa
Imperial College London, United Kingdom

Supervisor: Dr Robert Ewers
Microsoft Research supervisor: Drew Purves

Research title: A data-constrained predictive model of tropical deforestation and resultant carbon emissions

Research summary: Deforestation is a major source of global biodiversity loss and anthropogenic carbon emissions, but our ability to forecast the magnitude or geographical distribution of future deforestation is very limited at present. Making use of satellite-derived data sets measuring deforestation, in combination with global data on population density, climate and other factors, this project will (a) develop and parameterize a spatially explicit model of tropical deforestation; (b) combine this with data on carbon storage in tropical forests in order to estimate historical carbon loss from deforestation and to predict future losses.

John Hewson
University of Edinburgh, United Kingdom

Supervisor: Mr Paul Anderson
Microsoft Research supervisor: Rebecca Isaacs, Andy Gordon, Eno Thereska

Research title: Constraint-based specifications for system configuration

Research summary: This project aims to develop the use of constraint-based specifications for practical system configurations. Such specifications support composition which is particularly suitable for devolved management. They are also important for specifying autonomic systems in a declarative way since the under-specification provides the flexibility for autonomic adjustment. Reducing the complexity of the constraint problems, and handling ‘soft’ constraints are some of the important challenges to producing practical tools. The research direction will be guided by real problems in practical system administration, and it is expected to lead to tools or techniques that would be of use in this area.

Karen Simonyan
University of Oxford, United Kingdom

Supervisor: Prof. Andrew Zisserman
Microsoft Research supervisor: Antonio Criminisi

Research title: A medical image search engine

Research summary: We propose to develop a search engine for medical images that can return examples similar to a query image, e.g. with a particular anomaly, on the fly. We will harness modern and efficient methods of segmentation to provide the image descriptors, and visual words/a ‘Video Google’ architecture for the on-the-fly retrieval. The aim is to provide doctors with a retrieval tool to aid in their diagnosis by retrieving similar cases, their treatment and outcome. We will also investigate methods for data mining in collections of images, and improving segmentations given sets of aligned images or an image sequence.

Lars Schäfers
University of Paderborn, Germany

Supervisor: Prof. Dr. Marco Platzner
Microsoft Research supervisor: Thore Graepel, Satnam Singh

Research title: GOmputer: The GO machine

Research summary: The GOmputer project aims at the investigation of novel algorithmic approaches for playing GO and the development of a parallelized and hardware-accelerated GO machine prototype. In the algorithmic part of this project we will leverage recent Monte-Carlo approaches for playing GO and focus on a novel bundling technique, the combination of Monte-Carlo with alpha/beta game tree search, and the investigation of how to deal with patterns. In the computing systems part of this project, we will address suitable parallel programming models and advanced techniques for FPGA-based hardware acceleration. On the longer term, this project should lay the foundation for the development of the world’s strongest GO machine.

Lawrence Hudson
Imperial College London, United Kingdom

Supervisor: Dr Daniel Reuman
Microsoft Research supervisor: Lucas Joppa

Research title: Unifying food web structure and dynamics with metabolic theory: a general modular computational approach

Research summary: Theories of the dynamics of interacting species in food webs, though ubiquitous, are not sufficiently well connected with data to be useful for predictions. Much high quality static data on food webs and community structure is available, including recent data on allometry between species population densities and average body masses within a food web. The recent data shows regularities and systematic variation at a level of resolution suitable for precise tests of dynamical models. A flexible software package will be developed for constructing and parameterising dynamical models and for comparing their predictions with the new data to illuminate model mechanisms and constrain parameters. Best-fitting models will be used to make a variety of predictions about whole-food-web-level impacts of climate change, extinctions, and other disturbances.

Long Guo
Université d’Artois, France

Supervisor: Prof. Lakhdar Sais
Microsoft Research supervisor: Youssef Hamadi

Research title: Multicore-based satisfiability

Research summary: The SAT problem (decide if a Boolean formula, typically in conjunctive normal form, admits a valuation which makes it true?) is one of the fundamental problems in complexity theory, and probably one of the most studied, theoretically and practically. Modern SAT solvers can now handle propositional satisfiability problems with hundreds of thousands of variables or more. However, the whole picture is not so good since the SAT-solving community does only marginal performance gains (see results from the last SAT competitions). As a result, the progresses on industrial benchmarks are now stalling since it becomes harder and harder to improve the performances of any Chaff-like solver. Today, many SAT problems remain challenging to all the available SAT solvers, and consequently new approaches are clearly needed. In this context and with the light of the next generation of computer architecture, the design of multicore satisfiability solvers is clearly a fundamental issue. This is clearly a hot topic and research on this area is in a preliminary stage. The objective of this proposal is to provide new theoretical and practical advances for Multicore Satisfiability solving.

Martin Suda
Max Planck Institute for Software Systems

Supervisor: Prof. Christoph Weidenbach
Microsoft Research supervisor: Moritz Becker

Research title: Automated reasoning for dynamic authorization policy analysis

Analysing dynamic policies is challenging because it requires considering unbounded sequences of state changes. Recent work has proposed to leverage model checking tools, AI planners, and automated theorem provers, to analyse reachability and invariance properties of dynamic policies. However, the existing techniques are not entirely satisfactory for several reasons. The project will tackle some of these problems by using, extending or modifying an automated theorem prover such as Spass, or by developing and implementing new automated reasoning techniques.

Mindy Syfert
University of Cambridge, United Kingdom

Supervisor: Dr David Coomes
Microsoft Research supervisor: Matthew Smith

Research title: Novel computational approaches to modelling biodiversity – applications for setting conservation priorities

Research summary: The overall objectives of this project are to assess the accuracy of current methodologies for predicting the global occurrence of plant species based on presence data, and to investigate the development of potentially better alternative methodologies. It will develop new methods for establishing the climatic and edaphic constraints of species from presence-only data. It will compare separate techniques for the identification and proposal of protected areas. It will also test the coverage of the global protected area network for plants and identify additional priority areas for plant conservation.

Nadarajen Veerapen
Université d’Angers, France

Supervisor: Prof. Frédéric Saubion
Microsoft Research supervisor: Lucas Bordeaux

Research title: Autonomous neighbourhood management for combinatorial problems solving

Research summary: Neighborhood functions are crucial components when using local search algorithms to solve combinatorial optimization problems. Unfortunately, the design and the management of these functions are often problem dependant and require a great expertise and knowledge to obtain good results. Therefore, in order to provide more autonomous solving facilities, we propose to use genetic programming to generate suitable neighborhood functions together with suitable control features.

Nikée Groot
University of Sheffield, United Kingdom

Supervisor: Prof. Emanuel Gloor
Microsoft Research supervisor: Drew Purves

Research title: The other half of the equation: Global variation in tree mortality

Research summary: Future changes in the global forest carbon cycle could have major impacts on future climate. Whereas we now have a relatively advanced ability to model the input of carbon into forests, via tree growth, almost nothing is known about global variation in and / or the climate dependency of the outputs, which are dominated by tree mortality. By collating a unique, global forest inventory data set, and analysing this data using computational statistics, this project will, for the first time, uncover the global-scale climate dependency of tree mortality, and produce a model of this dependency for use in predictive models of the forest carbon cycle.

Sadia Ahmed
Imperial College London, United Kingdom

Supervisor: Dr Robert Ewers
Microsoft Research supervisor: Drew Purves

Research title: A data-constrained predictive model of tropical deforestation and resultant carbon emissions

Research summary: Deforestation is a major source of global biodiversity loss and anthropogenic carbon emissions, but our ability to forecast the magnitude or geographical distribution of future deforestation is very limited at present. Making use of satellite-derived data sets measuring deforestation, in combination with global data on population density, climate and other factors, this project will (a) develop and parameterize a spatially explicit model of tropical deforestation; (b) combine this with data on carbon storage in tropical forests, in order to estimate historical carbon loss from deforestation, and to predict future losses.

Sergey Kosov
Max Planck Institute for Software Systems, Germany

Supervisor: Dr Thorsten Thormählen
Microsoft Research supervisor: Pushmeet Kohli

Research title: From stereo to 3D faces

Research summary: The goal of this PhD project is to generate 3D reconstructions of faces using the two video streams provided by a stereo camera system. This involves automated detection, tracking, and recognition of faces and the extraction of certain properties. Compared to algorithms that use only a monocular camera, it is expected that the additional information provided by a stereo camera facilitates a more robust and accurate 3D reconstruction. A collection of real-time algorithms will be implemented that provide a basis for interactive applications, like, the augmentation of webcam videos, robots with stereo cameras for 3D vision, or the usage of a stereo camera as a new input device for games.

Su-Yang Yu
Newcastle University, United Kingdom

Supervisor: Dr Jeff Yan
Microsoft Research supervisor: Michael Roe, Ralf Herbrich

Research title: Cheating mitigation in online games

Research summary: While online game is a lucrative multibillion business, cheating in these games has become a serious issue for both game makers and players. In this proposal, I propose to tackle representative cheats in first-person shooters (FPS) and real-time strategy (RTS) games by developing novel security techniques, and also propose to explore the analysis of application programming interfaces (APIs) of Massively Multiplayer Online Games as a new means for identifying vulnerabilities in these systems. This project will contribute to both computer security and games research. Expected outcomes are not only of academic interests, but also of direct relevance to Microsoft’s gaming business.

Theofanis Karaletsos
Max Planck Institute for Biological Cybernetics, Germany

Supervisor: Dr Karsten Borgwardt
Microsoft Research supervisor: John Winn

Research title: Machine learning for genome-wide association studies and phenotype modelling

Research summary: The proposed thematic area for the MSR/MPI PhD candidate are machine learning challenges that arise in the context of genome-wide association studies (GWA). This covers general method development as well as the application of developed methods to real-world datasets. An overview of GWAs and related open problems in machine learning is given below.

Thomas Simpson
University of Cambridge, United Kingdom

Supervisor: Dr Alexander Duncan Oliver
Microsoft Research supervisor: Richard Harper

Research title: Trust on the Internet

Research summary: This project “seeks to understand how technology can support and enrich human values in the everyday world”, under the Socio-Digital Systems mandate, using philosophical methods. Trust and trustworthiness are of vital practical importance, without which human society cannot function. What are they, and how do they work? Further, the smooth functioning of the internet is critically dependent on well-placed trust. What are the specific challenges and opportunities that the internet presents? It is supposed that the values that are required for society to be marked by an atmosphere of trust are ‘thicker’ than liberal conceptions of what is required for us to be free from interference from others. The construction of a future where technology facilitates and enhances the reliable operation of our trust habits is the invention of a future where these values are strengthened, rather than eroded. Thus there is a close iterative relationship between technology and values, which this project seeks to explore.

Timothy Rudge
University of Cambridge, United Kingdom

Supervisor: Dr James Haseloff
Microsoft Research supervisor: Andrew Phillips

Research title: Language for synthetic biology

Research summary: The overall objective of the project will be to develop a high-level programming Language for Synthetic Biology (LSB) that will allow formal representation of synthetic designs, simulation and possible solutions. The student will apply Synthetic Biology tools to a bacterial system to generate self-organizing “Turing” patterns that can be formally modelled, generating essential information for the development of LSB. The development of an improved theoretical tool like LSB requires use of a practical biological system for testing and refining the language. Here the student will use an existing bacterial genetic system, the Gram-positive bacterium Bacillus subtilis. The student will benefit from collaboration with an existing programme in the Ajioka and Haseloff laboratories at the University of Cambridge, and participate in the design and construction of a novel library of standard biological parts (known as G+Bricks). These DNA parts will include promoters, terminators, ribosome binding sites and protein coding sequences for use in Bacillus. With these parts, the student will be able to test biological circuits for cell-cell communication. Diffusible signals can work as positive and negative regulators to determine the state of individual cells, establishing local conditions within a population, and form self-organised spatio-temporal patterns, such as those described by Turing. The theoretical development and practical testing of LSB will be tied to the construction of simple self-organising circuits in Bacillus subtilis. These patterning devices will have potential wide application as intelligent switches in genetic systems.

Vijay D’Silva
Oxford University, United Kingdom

Supervisor: Dr Daniel Kröning
Microsoft Research supervisor: Josh Berdine

Research title: Generalisation operators for abstraction-refinement

Research summary: Abstraction-refinement with predicates forms the basis of state-of-the art software verification tools. Unfortunately, many existing abstraction-refinement techniques diverge even on extremely simple examples, for which finite abstractions are known to exist. The aim of this proposal is to study and identify the source of divergence in abstraction-refinement methods and combat them using generalisation operators. Such operators are used to identify and generalise a pattern in a sequence of abstractions. Unlike abstraction, the basic principles behind the design and analysis of generalisation operators are not well understood. We propose designing generalisation operators using learning algorithms and interpolation-based methods. On the theoretical side, we will study the completeness and termination properties of abstraction-refinement using such operators. On the practical side, we will develop and evaluate a tool for software verification based on abstraction-refinement with generalisation operators.

Ye Yuan
University of Cambridge, United Kingdom

Supervisor: Dr Jorge Goncalves
Microsoft Research supervisor: Hillel Kugler

Research title: Robust network reconstruction with applications to biology

Research summary: This project focus on the development of mathematical and software tools to reconstruct causal network structures from data and with application to regulatory T (Treg) cells. These results will allow biologists to unveil causal network structures between measured species. With time series data, this network gives computational predictability through simulations where hypothesis can be tested without the need to perform new experiments. With steady-state data, in addition to the Boolean network, it reveals the gains between measured species and whether they activate or inhibit each other. The discovery of the network helps understand how the system functions. This is fundamental if we wish to control it with synthetic biology to produce a desired system or for drug development. Knowing the actual causal network structure allows to change or remove a specific pathway without affecting others. Given noise present in the data and nonlinearities in the system, we will find the most likely candidate networks that fit the data by measuring the smallest distance, in some norm, from the data to all Boolean network structures. On the application side, we will reveal the network structure among a carefully selected set of genes of Treg cells in collaboration with Professor Balling from the Helmholtz Center for Infection Research in Germany, where all the experiments will be performed. The obtained network structure will help us understand the complicated dynamic functional mechanism of Treg cells and further help produce therapeutic strategies of many of autoimmune diseases in which Treg cells play a vital role.