CSC PhD Studentships in Electronic Engineering and Computer Science
Course: PhDs in Electronic Engineering and Computer Science
Value: Full tuition fee waiver and living stipend (£1350/month) for 4 years
Deadline: 31st January 2023
About the studentships
The school of Electronic Engineering and Computer Science of the Queen Mary University of London is inviting applications for several PhD Studentships in specific areas in Electronic Engineering and Computer Science (please see at the end of this page for a list of projects) co-funded by the China Scholarship Council (CSC). CSC is offering a monthly stipend to cover living expenses and QMUL is waving fees and hosting the student. These scholarships are available only for Chinese candidates.
About the School of Electronic Engineering and Computer Science at Queen Mary
The PhD Studentship will be based in the School of Electronic Engineering and Computer Science (EECS) at Queen Mary University of London. As a multidisciplinary School, we are well known for our pioneering research and pride ourselves on our world-class projects. We are 8th in the UK for computer science research (REF 2021) and 7th in the UK for engineering research (REF 2021). The School is a dynamic community of approximately 350 PhD students and 80 research assistants working on research centred around a number of research groups in several areas, including Antennas and Electromagnetics, Computing and Data Science, Communication Systems, Computer Vision, Cognitive Science, Digital Music, Games and AI, Multimedia and Vision, Networks, Risk and Information Management, Robotics and Theory.
For further information about research in the school of Electronic Engineering and Computer Science, please visit: http://eecs.qmul.ac.uk/research/.
Who can apply?
Queen Mary is on the lookout for the best and brightest students. A typical successful candidate:
- Should hold, or is expected to obtain an MSc in the Electronic Engineering, Computer Science, or a closely related discipline
- Having obtained distinction or first-class level degree is highly desirable
Eligibility criteria and details of the scheme
How to apply
Please see at the end of this page a list of potential PhD projects and supervisors.
Applicants should work with their prospective supervisor and submit their application following the instructions at: http://eecs.qmul.ac.uk/phd/how-to-apply/
The application should include the following:
- CV (max 2 pages)
- Research proposal (max 500 words)
- 2 References
- Certificate of English Language (for students whose first language is not English)
- Other Certificates
The deadline for applications is the 31st of January 2023.
For general enquiries contact Mrs. Melissa Yeo email@example.com (administrative enquiries) or Professor Ioannis Patras firstname.lastname@example.org (academic enquiries) with the subject “EECS 2023 CSC PhD scholarships enquiry”.
List of available projects and corresponding academics:
Supervisor: Dr Ahmed Sayed
AI/ML systems are becoming an integral part of user products and applications as well as the main revenue driver for most organizations. This resulted in shifting the focus to bringing the intelligence towards where the data are produced including training the models on these data. Existing approaches operate as follows: 1) the data is collected on multiple servers and processed in parallel (e.g., Distributed Data-Parallel); 2) the server coordinates the training rounds and collects model updates from the clients (e.g., Federated Learning); 3) the server splits the model training between the clients and the server (e.g., Split Learning); or 4) the clients coordinate among themselves via gossip protocols (i.e., Decentralized Training). The challenges that manifest themselves are the highly heterogeneous learners, configurations, and environment, communication and synchronization overheads, fairness and bias, and privacy and security. Therefore, existing approaches fail to scale with a large number of learners and produce models with low qualities and high bias at prolonged training times. It is imperative to build systems that provide high-quality models in a timely manner. This project addresses this gap by exploring novel ideas and proposing efficient and scalable ML systems for decentralized data.
Supervisor: Dr Ahmed Sayed
Despite the huge success of the Internet, in many scenarios, the connectivity often falls short of expectations leading to devastating impacts on our services and applications. Currently, the legacy networks, which are the nerve system connecting all our services, lack sufficient intelligence to support the fast-paced evolution of many applications. This is a major obstacle that will be exacerbated when advanced applications like virtual reality, metaverse, autonomous vehicles, smart cities, digital economy, and real-time healthcare wish to be successfully deployed. Addressing this concern requires a drastic departure from legacy hard-coded and human-error-borne networking systems to developing more intelligent, agile and responsive networking systems. Thanks to the latest breakthroughs in software-defined networking, hardware programmability and ML for systems, in this project, we will tap into these developments to seek solutions to the current problems in a way that is deployable incrementally. This project will explore novel ML-based networking solutions and develop an ML-based support framework as the basis for Future Intelligent Networks (FIN).
Supervisor: Dr Anthony Constantinou
Discovering accurate cause-and-effect relationships from data represents a highly challenging task. Causal ML algorithms tend to be evaluated with clean synthetic data that conform to the ideal learning assumptions of the algorithm under assessment. Because even minor imperfections in the input data would violate multiple of the learning assumptions of an algorithm, applying these algorithms to real data - which is almost always imperfect in different ways - makes the task of causal discovery notoriously difficult.
This project will focus on devising new causal ML algorithms that can more effectively deal with imperfect data. The aim is to optimise learning for data imperfections found in practice, such as missing and biased data, rather than assuming the input data are error-free and faithful to a set of ideal algorithmic assumptions.
In turn, this project will lead to guidelines for when causal ML might be appropriate, and what level of accuracy one might expect from a given algorithm and imperfect data combination.
Supervisor: Dr Anthony Constantinou
Assessing and modelling causal relationships is fundamental to identifying and explaining the causal phenomena we experience. For example, government policy and hypothesis-based clinical research largely involve weak assumptions of causal structure that enable decision making through intervention. Causal Machine Learning (ML) represents the field of ML that focuses on discovering causal models from data. These models are intended for application to real-world problems that go beyond prediction and towards optimal decision making and control via intervention.This project will explore ML approaches that enable us to extract as much causal information as possible from real-world datasets, such as healthcare datasets that are imperfect in different ways, and using that information in the most efficient way for causal representation and causal inference.
Supervisor: Prof Greg Slabaugh
Neural Rendering Fields (NeRFs) takes a collection of input photographs as input. The method produces a neural network model that produces photorealistic views of the scene rendered from new viewpoints. We are interested in applying NeRF to humans to produce avatars, aka digital humans.
We recently published work at BMVC 2022 that “disentangles” NeRF methods to give the user control of not only the viewpoint, but also the pose and clothing of the avatar. This is a great step forward, however the current state-of-the-art is limited in its ability to realistically capture fine geometric details like hands and crisp visual details around the face. Often these regions appear blurry when rendered using NeRF.
In this research we will extend this work to better model the full body including detailed representations of the hands and face. This will improve the ability of NeRF to capture humans and enables new methods for people to engage with their photographs using advanced computational photography techniques. We will also explore how to reduce the number of input photographs to make the scene easier to capture.
This class of methods has numerous applications, for example in smartphone photography, computer games, digital video special effects, and healthcare.
Supervisor: Dr Haim Dubossarsky
The ability of Language Models (LM) to capture essential linguistic features, whether it is syntactic, semantic, or others, remains largely mysterious. This is in part because the tools currently used to investigate LM are too basic to analyze the intricate geometry of the neural representations (or embeddings) produced by deep neural networks. This studentship will re-think the way we analyze embedding spaces today, and would employ and develop methods from Topological Data Analysis (TDA), which is a collection of data-driven methods based on the mathematical field of algebraic topology. TDA provides simplified representations of complex data, which is ideal for our challenging problem, and is commonly used to extract topological features underlying high dimensional representations. The studentship will include the development of new TDA-based measures to better describe embedding spaces and the information they encode, and will extend the prior work of the supervisors in this area (Dubossarsky et al., 2020; Meirom & Bobrowski, 2022).
This PhD will be supervised by Dr Haim Dubossarsky & Dr Omer Bobrowski. For more information, please contact Dr Dubossarsky.
The student will be based in the Cognitive Science Group at the School of Electronic Engineering and Computer Science, as well as in the Statistics and Data Science Group at the School of Mathematical Sciences.
Dubossarsky, H., Vulić, I., Reichart, R., & Korhonen, A. (2020). The Secret is in the Spectra: Predicting Cross-lingual Task Performance with Spectral Similarity Measures. EMNLP, 2377–2390.
Haim Meirom, S., & Bobrowski, O. (2022). Unsupervised Geometric and Topological Approaches for Cross-Lingual Sentence Representation and Comparison. 173–183. In RepL4NLP.
Supervisor: Prof Matthew Purver
Supervisor: Dr Ignacio Castro
Second Supervisor: Dr Richard Clegg
Social networks are ubiquitous in modern life, they are a fun way to interact with friends as well as read news and pass on information. “Echo chambers”, are a well-known problem of social networks that reinforce the opinions of its users and lead to polarisation.
This project will look at the emergence and evolution of echo chambers and in particular at how new “ideas” and “topics” emerge and evolve and circulate between different echo chambers. This is a data-driven project that could be approached from a computational social-science and/or mathematical angle (depending on student preference). The project might use Machine Learning, Social Network Analysis, and/or Natural Language Processing techniques. There is flexibility to shape the exact nature of the project around this core idea and the student will work with the supervisors to agree on the exact goals and methods.
Supervisor: Dr Shady Gadoue
Multiphase electric drive powertrains are becoming increasingly popular in marine and automotive applications. The multiphase drive powertrain system has many features including increased reliability and efficiency, fault-tolerant operation, and reduced converter rating. Online battery state estimation techniques are required to continuously monitor the battery state of health and state of charge during operation and further predict failure and ageing of the battery system. No much research has been carried out on online state estimation of battery systems in those automotive and marine propulsion applications employing multi-phase drive systems.
The use of existing estimation techniques, based on the conventional drive powertrain system, affects the system performance including induced unwanted torque ripples. Due to the features of extra degrees of freedom offered by the proposed multi-phase configuration, it is expected to reduce the impact on the system performance. This project will investigate the use of the multi-phase power converter modulation to produce the battery impedance graphs online and, in conjunction with predictive control techniques, ageing models for the battery system will be developed, all achieved with minimal effect on the powertrain drive performance. Another feature of the proposed technique is the utilisation of the already-existing power converter for the production of the perturbation signal required for the prediction; thus reducing the cost and size of the estimation process, a feature greatly required in on-board electric automotive and marine propulsion systems.
Supervisor: Prof Sean Gong
Object recognition in computer vision requires a model to perform two simultaneous tasks: localisation and classification. Modern deep learning object recognition models are poor on small and/or occluded objects. Robust object recognition is challenging to data-centric machine learning. Data augmentation is a technique for improving model learning under unknown biased data sampling. For objects with small training data of unknown bias in sampling, distribution expansion has been shown to be effective. This research will investigate self-learning methods with distribution expansion for learning unbiased object recognition.
The student will be based in the Computer Vision Group at the School of Electronic Engineering and Computer Science.
1. J. Huang, S. Gong. “Deep Clustering by Semantic Contrastive Learning”. In Proc. British Machine Vision Conference, London, November 2022.
2. Q. Li, J. Huang, J. Hu, S. Gong. “Feature-Distribution Perturbation and Calibration for Generalized Person ReID”. arXiv preprint arXiv:2205.11197, May 2022.
3. P. Li, D. Li, W. Li, S. Gong, Y. Fu, T. Hospedales. “A Simple Feature Augmentation for Domain Generalisation”. In Proc. IEEE International Conference on Computer Vision, Montreal, Canada, October 2021.
Supervisor: Dr Mahdieh (Maddie) Sadabadi
The increasing deployment of power electronics converters in safety-critical applications (such as microgrids and electric vehicle charging stations) demands robustness against system faults and deteriorations. The complexity of the dynamics of power electronics converters, along with the major investment in process data collections, have highlighted the need for data-driven approaches for diagnosis and prognostics, as a fundamental tool for predictive maintenance mechanisms in such systems.
The main aim of this project is to develop a health-aware control system for power electronics converters that integrates the information provided by a data-driven prognosis module about converters’ health to accordingly adapt the controller, extending the life of the power electronics converters. The developed health-aware control technologies will be verified by real-time digital simulators and hardware-in-the-loop experimental testbeds.
Prospective applicants should have a first-class degree and/or Master’s degree (Distinction) or be close to completion in Control Engineering/Electrical Engineering or related subjects. In addition, you should also have a strong background in control systems. Familiarity with machine learning and power electronics is a plus.