- Andrej Mikulik, Czech Technical University, Czech Republic
Large scale image search for objects and categories - Connie Golsteijn, University of Surrey, United Kingdom
Materialising media - Danielle Belgrave, Dorothy Hodgkin Postgraduate Award, University of Manchester, United Kingdom
Probabilistic causal models for asthma and allergies developing in childhood - David Silver, Technion, Israel
A systems biology approach to detect adverse patient-drug interactions - Gian Marco Palamara, University of Zurich, Switzerland
Computational modelling of the collapse of ecological communities - Ismail Kuru, Koç University, Turkey
A static proof system and tool for programs running on relaxed memory models - Lara Houston, Lancaster University, United Kingdom
Inventive infrastructures—an exploration of mobile phone ‘repair’ cultures in Uganda - Larissa Pschetz, Dorothy Hodgkin Postgraduate Award, University of Dundee, United Kingdom
Are we nearly there yet? A proposal to explore digital navigation - Michal Ficek, Czech Technical University, Czech Republic
Understanding and modelling network migration - Nicolas Mobilia, CNRS, France
A meta environment for biological network modelling - Niek Bouman, Eindhoven University of Technology, The Netherlands
Distributed spectrum sharing for wireless networks: Optimal performance, fairness and design - Paul Kelly, University of Oxford, United Kingdom
Assessing the potential for SenseCam to fight the current global public health crisis of increasing obesity and physical inactivity - Peter Wortmann, University of Leeds, United Kingdom
Visualising performance for multi-core Haskell - Pravin Shinde, ETH Zurich, Switzerland
Scalable networking for heterogeneous multi-core systems - Varun Bhaskar Kothamachu, Dorothy Hodgkin Postgraduate Award, University of Exeter, United Kingdom
Computational capabilities and underlying mechanisms in biological signalling networks - Volodymyr Kuznetsov, ÉPFL, Switzerland
Selective symbolic execution
Andrej Mikulik
Czech Technical University, Czech Republic
Supervisor: Dr Jiří Matas
Microsoft Research supervisor: Andrew Fitzgibbon
Research title: Large scale image search for objects and categories
Research summary: This PhD project will investigate large scale image search techniques, their improvements for specific object retrieval, broadening their applicability and extending those approaches to object class search.
Connie GolsteijnUniversity of Surrey, United Kingdom
Supervisors: Prof. David Frohlich, University of Surrey; Dr. Elise van den Hoven, Technical University of Eindhoven
Microsoft Research supervisor: Abigail Sellen
Research title: Materialising media
Research summary: Digital media are currently revolutionising the way we capture and share personal experiences through photos, sounds, video and other types of personal content. This change has been most pronounced in the area of domestic photography where digital cameras, camera-phones and internet services have come to replace analogue cameras and photographic prints as the primary means of capturing and sharing snapshots. This has led to the exponential growth of digital photo collections, alongside existing print archives. Similar changes are happening in consumer video and music, leading to a bifurcation of old and new technology and their associated media.
Dorothy Hodgkin Postgraduate Award, University of Manchester, United Kingdom
Supervisor: Prof. Iain Buchan
Microsoft Research supervisor: Christopher Bishop
Research title: Probabilistic causal models for asthma and allergies developing in childhood
Research summary: This trans-disciplinary PhD will focus on the exploitation of Bayesian machine learning methods, based on probabilistic graphical models, in the quest to understand the determinants of asthma and allergies from childhood, including the interactions between genetic and carefully-measured environmental factors. Structured Bayesian models will be built and solved using the Infer.NET library, and will be evaluated alongside conventional bio-statistical methods, such as multi-level models. The ultimate goal of the project is to elucidate realistically-complex causal networks of genetic and environmental factors responsible for asthma and allergies that develop in childhood. An allied PhD project, funded by the NIHR, will take a graphical model approach to studying the factors that determine the outcomes of treating type 2 diabetes. This will use the same type of genetic data and Infer.NET.
David Silver
Technion, Israel
Supervisor: Dr. Itai Yanai
Microsoft Research supervisor: Hillel Kugler
Research title: A systems biology approach to detect adverse patient-drug interactions
Research summary: Progress in genomic research has enabled the identification of interactions among the genome and the environment; however, the medical sciences still await the translation to a general method for predicting the adverse side-effects of therapeutics. In part, the delay is due to difficulties involved in attaining a dataset with the required depth and breadth to allow the formulation of a rigorous and general method. Here we propose to address this problem in the experimentally tractable nematode C. elegans where an adequate dataset can be constructed, the underlying principles discovered, and the tools then translated to sparse human datasets. Our method is premised on the notions of gene expression as an efficient marker for disease as well as its overall modularity in co-expressed subsets, and involves the statistical classification and unsupervised learning of large-scale datasets. We hypothesize that an interaction may be inferred from the expression levels of just a small set of genes and conditions (drugs), based upon knowledge of the underlying correlations of expression among the entire set of genes. We propose that this is an efficient approach towards discovering the principles of patient/drug interactions which may contribute to a physician’s ability to predict with reasonable confidence the possible adverse effects of a range of possible therapies.
Gian Marco PalamaraUniversity of Zurich, Switzerland
Supervisor: Dr. Owen Petchey
Microsoft Research supervisor: Matthew Smith
Research title: Computational modelling of the collapse of ecological communities
Research summary: We propose a research project about how and why species’ traits, inter-specific dependencies, and chance, influence the trajectory of ecosystem failure that results from species extinctions. As well as providing a step forward in the understanding of the drivers of ecosystem failures and the ability to provide more realistic extinction scenarios, we imagine findings will apply to other types of systems in other scientific fields. In particular, systems in which the failure of one component can affect the failure risk of other components, a situation that may have caused some of the most significant disasters in recent history.
Ismail Kuru
Koç University, Turkey
Supervisor: Serdar Tasiran
Microsoft Research supervisor: Shaz Qadeer
Research title: A static proof system and tool for programs running on relaxed memory models
Research summary: We propose a static proof system and an associated mechanical proof checker for programs running on relaxed (weak) memory models. We target the verification of highly-concurrent, highly-optimized code such as virtual machines, language runtimes, transactional memory implementations, and libraries providing concurrency primitives and concurrent data structures. All applications that use such infrastructure software rely on its correctness. We believe that bug-finding tools are insufficient for ensuring this critical correctness. The use of more heavyweight verification approaches is justified for concurrent infrastructure software, and mechanically-checked proofs are necessary.
We have previously developed QED, an atomicity, abstraction- and reduction-based proof system and tool, and had success proving the correctness of intricate concurrent code using QED. Unlike application software, the highly-concurrent low-level code on which this proposal focuses does not first ensure and then rely on sequential consistency for correctness. This has led us to propose a generalization of the QED proof system and tool in order to take into account relaxed memory models.
Lara Houston
Lancaster University, United Kingdom
Supervisors: Prof. Lucy Suchman; Dr. Adrian Mackenzie
Microsoft Research supervisor: Alex Taylor
Research title: Inventive infrastructures – An exploration of mobile phone ‘repair’ cultures in Uganda
Research summary: This research aims to address two general areas. The first focuses specifically on the practices of use of mobile phones in Uganda, and aims to contribute to an emergent body of work around mobile telephony and ICT in developing regions (the latter somewhat troublingly referred to as M4D and ICTD). The second area of focus will investigate local practices of mobile phone repair and maintenance, and will consider what if any lessons can be learned for ICT in Kampala and more generally.
Larissa Pschetz
Dorothy Hodgkin Postgraduate Award, University of Dundee, United Kingdom
Supervisors: Dr Jon Rogers, University of Dundee; Dr Chris Speed, Edinburgh College of Art
Microsoft Research supervisor: Richard Banks
Research title: Are we nearly there yet? A proposal to explore digital navigation.
Research summary: The way we navigate has changed, and so too has the way in which we understand the context and environments: the roads, streets, and highways across which we travel. Digital technology is starting to alter the way we plan to get about, how we navigate getting about and the way we get back from getting about. From personal, to private, to public, to group travel—we are starting to do this differently and this may be the start of a cultural shift in the business of navigation. Understanding the social implications for these technologies is vital if we are to design better products to help people navigate the landscapes of the present and the future. We aim to investigate the future of navigation that focuses on the domestic routines of people in their homes, their neighbourhoods, in their relationships, and in their landscapes.
Michal Ficek
Czech Technical University, Czech Republic
Supervisor: Dr Lukas Kencl
Microsoft Research supervisor: Milan Vojnovic
Research title: Understanding and modelling network migration
Research summary: Contemporary wireless communication networks present many alternatives for network users to obtain access to the Internet and mobile telephony connectivity. The association of a user mobile terminal to a particular access network may change over time depending on his or her country of location (cross-border roaming), availability of a network signal, pricing plan, provider preference, time of day, location, device capability or desired application of use (voice, data, etc). In this work we propose to conduct research in designing, building and applying tools to trace, analyse, measure and model such network migration behaviour of user terminals and determining the general characteristics of network migration in relation to the space and time attributes. Furthermore, we propose to formulate general guidelines for designing various network services or functions such as content distribution, virtual operators, energy-efficient performance or roaming-customer retention, based on the above findings.
Nicolas Mobilia
CNRS, France
Supervisors: Dr Éric Fanchon, CNRS; Prof. Jacques Demongeot, Université Joseph Fourier
Microsoft Research supervisor: Youssef Hamadi
Research title: A meta environment for biological network modelling
Research summary: Formal modelling is becoming an essential part of the work of today’s biologists. Our global goal is to develop software tools based on constraint technologies to assist biologists in the process of building models of complex interaction networks. The benefits of such system-level models are: deepening our understanding of life (at the molecular and cellular levels in our case), design of combinatorial therapies taking into account the network structure and exploiting fragility points. The usual process of knowledge acquisition can be schematically viewed as a succession of data production phases and modelling phases. Knowledge being partial, families of models should be considered rather than single instantiated models. A family of models embodies part of the state of knowledge at a given time, and often incorporates hypotheses in addition to features supported by experimental data. A family of model thus defines a synthetic reasoning frame. We advocate here a rational approach to network modelling based on constraints.
Niek Bouman
Eindhoven University of Technology, The Netherlands
Supervisors: Prof. Sem Borst; Dr. Johan Van Leeuwaarden
Microsoft Research supervisors: Peter Key; Alexandre Proutiere
Research title: Distributed spectrum sharing for wireless networks: Optimal performance, fairness and design
Research summary: The proposed research focuses on distributed spectrum sharing algorithms for emerging large-scale wireless meshes and cognitive-radio networks. In contrast to today’s cellular architectures, these networks typically lack any centralized control entity for allocating resources and explicitly coordinating transmissions. Instead, these networks vitally depend on the individual nodes to operate autonomously and efficiently share the medium in a distributed fashion. This requires nodes to schedule their individual transmissions and decide on the use of shared resources based on knowledge that is locally available or only involves limited exchange of information. The paradigm of such distributed control has been successfully adopted for end-to-end congestion control in wired communication networks. Wireless networks, however, have fundamentally different characteristics and entail even bigger challenges, particularly due to unpredictable channel conditions and complex interference issues. In cognitive-radio environments, a further key requirement is that the opportunistic access by secondary (unlicensed) devices of unused portions of the spectrum (white spaces) must not interfere with the transmissions of primary users (incumbents). This raises a strong need for agile spectrum sensing techniques and intelligent probing algorithms in order to access the spectrum in an efficient and non-intrusive fashion.
Paul KellyUniversity of Oxford, United Kingdom
Supervisor: Dr. Charlie Foster
Microsoft Research supervisors: Steve Hodges; Emma Berry
Research title: Assessing the potential for SenseCam to fight the current global public health crisis of increasing obesity and physical inactivity
Research summary: This proposal outlines the potential role of SenseCam device in public health research and understanding. Based on our small pilot study we have identified and tested the potential for using SenseCam to validate the existing benchmark national measure of active travel, used on the National Travel Survey (NTS); indeed our research partners in this application NatCen, who conduct the National Travel Survey, have agreed to field test SenseCam in principle, subject to the development of appropriate protocols. We are confident that SenseCam contains a unique blend of methods from criterion and objective assessment of physical activity. The instrument can directly observe and record, through the camera mechanism, the duration and type of physical activity that is being performed, it’s place and context and provide a criterion method to assess the reliability and validity of a self report methods, for example a diary or questionnaire. SenseCam is an improved, more accurate way to measure physical activity behaviour, and the information it gives us will allow for better understanding of the determinants of behaviour and thus the design of better interventions to promote physical activity.
Peter Wortmann
University of Leeds, United Kingdom
Supervisor: Dr. David Duke
Microsoft Research supervisor: Simon Peyton-Jones
Research title: Visualising performance for multi-core Haskell
Research summary: The cost and complexity of programming multi-core processors is encouraging industrial interest in pure functional programming. But high-level abstractions make it difficult to isolate and resolve performance problems. Visualization of run-time behaviour could help, but existing methods were designed for lower-level technologies. This project will investigate how to provide Haskell programmers with salient source and run-time information that underpins a reasoned approach to performance tuning.
Pravin Shinde
ETH Zurich, Switzerland
Supervisors: Prof. Timothy Roscoe; Prof. Gustavo Alonso
Microsoft Research supervisor: Paul Barham
Research title: Scalable networking for heterogeneous multi-core systems
Research summary: This project will look at whether it is beneficial to rethink the architecture of the OS network stack in the light of new hardware, both network interfaces and overall system architecture. The key idea is to take a global system view, in particular with regard to dynamically placing elements of the networking stack appropriately in the end system. We will revisit the traditional idea of modelling the stack as a dataflow graph of protocol elements, but with significant differences.
Varun Bhaskar Kothamachu
Dorothy Hodgkin Postgraduate Award, University of Exeter, United Kingdom
Supervisor: Dr. Orkun Soyer
Microsoft Research supervisor: Luca Cardelli
Research title: Computational capabilities and underlying mechanisms in biological signalling networks
Research summary: There are still major challenges to overcome before we can claim to understand key “design” properties underlying signalling networks. Firstly, it is not clear how features identified so far will pre-dispose certain response dynamics in a larger network where they are embedded. For example, theory suggests that same signalling elements can easily result in different response dynamics when connected in different ways. Secondly, the conclusions of most modelling studies are found to depend on the details of the model structure. In particular, molecular events that are usually disregarded in models as being negligible can significantly influence response dynamics. Finally, and most importantly, we lack detailed understanding of how evolutionary processes can move signalling systems in the “space” of possible response dynamics. In other words, we do not fully understand how mutations can generate novel response dynamics from existing ones. The main aim of the proposed research is to address these challenges by systematically and exhaustively analyzing topological and biochemical features in tractable signalling systems.
Volodymyr Kuznetsov
ÉPFL, Switzerland
Supervisor: Prof. George Candea
Microsoft Research supervisor: Manuel Costa
Research title: Selective symbolic execution
Research summary: Our goal is to develop an automated testing technique that can scale to millions of lines of code that interact with their environment. We build upon symbolic execution, a technique originally introduced in the 1970s that has recently gained popularity in automated software testing. Behaviours (such as bugs) discovered with symbolic execution can be easily reproduced using information collected during the corresponding symbolic execution, making this approach a powerful and cost-effective tool for developers and testers.
Meet the PhD Scholars
|
|