Electronic Engineering & Computer Science

Below you will find Data-Centric Engineering projects offered within the School of Electronic Engineering & Computer Science

This is not an exhaustive list. If you have your own research idea, or if you are a prospective Professional Doctoral Scholar candidate, please return to the main DCE Research webpage for further guidance, or contact us at dce-cdt@qmul.ac.uk 

Causal AI using EHR data for clinical decision support

The digitisation of electronic health records (EHRs) has opened up opportunities to analyse and leverage data at scale for the purpose of supporting health professionals to deliver effective and efficient care. In particular there is high interest in targeting resources for best patient outcomes. To achieve this, we need algorithms that can handle large amounts of heterogeneous data. For instance the format of EHR data is heterogeneous: both structured (e.g. clinical codes, prescriptions) and unstructured (e.g. free-text letters), at different levels of detail (e.g. ward notes vs discharge summaries), for different purposes (referral, diagnosis, billing, etc.), and by different sets of actors (doctors, nurses, physiotherapists, etc.). Further, EHR data is heterogeneous in terms of topic, describing different aspects of a patient’s care (e.g. medications, diagnoses, treatments, social factors), for different medical specialties (e.g. cardiology, psychiatry, dermatology) and in different settings (e.g. primary care vs secondary care). Algorithms require understanding of the interplay between factors, including the causal relationships, in order to model the data in a way that enables intervention at an individual or system level.

This project will leverage large EHR datasets available in the public domain and via Canon clinical collaborators, build machine learning models for characterising and stratifying patients. The focus will be on two lines of investigation:

• developing general medical machine learning methods that can be trained using supervision from existing data, with minimal additional expert input e.g. by cross-referencing different documents and data types, or by using clinical coding as supervision to extract further structure from free-text data.
• exploring methods of integrating understanding of causality such that machine learning models can provide accurate explanations of patient status and predictions for future outcomes given potential interventions.

Supervisors: Dr Alison O'Neil (Principal Scientist, Canon Medical) and Prof Maria Liakata

Please note that this project is being offered in partnership with Canon Medical Research Europe, based in Edinburgh.  The successful candidate can be based in Edinburgh for the duration of the studentship, with regular visits to the QMUL Mile End campus.  If the candidate prefers to be mainly based elsewhere, it will be possible to negotiate a combination of remote work with regular visits to Edinburgh and QMUL.

Resource Efficiency in Federated Learning EcoSystems

In this project, we explore the promising prospects and imminent challenges towards the facilitation of the adoption of federated learning for service providers. The prospects are motivated by the growing interest and momentum towards the adoption of 5G/6G technologies. Whereas, the challenges that hinder wide FL adoption are mainly the resource and user heterogeneity, and communication overhead which remains unaddressed. In this project, we aim to follow a scientific and practical research methodology in pursuit of addressing these challenges. We also aim to leverage the huge industrial interest into federated learning to build a collaborative framework with key industrial partners.

Supervisor: Dr Ahmed M. Abdelmoniem

Data-Driven Intelligent Audio Technologies

Many audio technologies require expert users in order to be properly applied. Dynamic range compression, room correction, acoustic feedback prevention etc. all use established signal processing technologies, with limitations. Either they need manual fine-tuning or are usable only in limited situations. Modern machine learning techniques may hold the key to automatically and intelligently applying parameters. Neural networks may be trained to optimise settings for different conditions.

The project is concerned with investigating, implementing and evaluating intelligent systems that render audio content towards a target. It will explore new approaches to challenging problems, with wide application to industry and potential end users.

The first stage of work involves a preparatory literature study, defining use cases and developing an initial prototype. The next stage will involve working with and learning from existing data sets and experimental measurements. The final stage is concerned with objective and subjective evaluation, as well as development of metrics and automated correcting of system parameters from subjective evaluation.

The project can be shaped to the researcher’s interests and expertise. It is expected that the research could yield high impact publications and have application in wider machine learning contexts.

Supervisor: Prof Josh Reiss 

Composition-aware music recommendation systems for music production

Music recommender systems (MRSs) have to date primarily focused on song or playlist recommendation for listening purposes, but much less work has been done on the recommendation of audio content for music production. Contemporary music production tools commonly include large digital audio libraries of loops (repeatable musical parts typically used in grid-based music), virtual instrument samples (based on synthesis or recordings) and sound packs (collections of sounds). Such richness of content can however impede creativity as it can be daunting for musicians to navigate tens of thousands of sounds to find items matching the style of their production and intent.

This project will research, develop and assess composition-aware music recommendation systems for music production enabling musicians get the best musical value out of their creative digital audio libraries.

Computer music makers often produce music by combining different instrumental parts together (e.g. drums, bass, lead, vocals, etc.). One of the challenges will be to provide new audio items that can be meaningfully mixed with other audio elements in the composition. The project will investigate methods to assess the musical compatibility between a set of audio items given constraints such as a composer’s musical taste and creative style.

Different music recommendation paradigms and AI techniques will be researched in order to minimise the impact of cold-start and sparsity issues, and maximise interpretability of the results. The project will compare content-based filtering (CBF) using audio or metadata (e.g. “analog dirty bass”, “crisp bright lead”, etc.), collaborative filtering (CF) based on user-item interactions, and hybrid models. Graph-based solutions for composition-aware recommendation will be investigated, e.g. by developing multi-layer structure taking into account user preferences and audio item musical compatibility. The project will also research deep learning techniques for (i) automatic feature learning from audio (e.g. to model perceptual distances between sounds to find sound-alikes), (ii) modelling audio item musical compatibility, and (iii) extracting latent factors from user-item interactions.

Evaluations of the models will be conducted taking into account how they support creative agency and provide interpretable recommendations to the user.

The scholar will have access to data and software provided by Focusrite including a catalogue of over 10k WAV loops with metadata, a collection of about 30k synth patches (with about 200 parameters), and musical compatibility indicators. The recommendation models will be tested on Ampify iOS music apps such as Launchpad, to make and remix music, and Blocs Wave, to make and record music  (https://ampifymusic.com/).

Supervisor: Dr Mathieu Barthet

Data-Centric Engineering methodology to enable emerging wireless technologies

This project aims to develop novel data-centric methodology and traceable techniques informed by machine learning and real data capture to enable emerging wireless technologies – e.g. Internet of Things (IoT), fifth-generation (5G) and sixth-generation (6G) mobile networks, to improve measurement efficiency and uncertainties to underpin all aspects from the systems, environments, and exposures and to provide metrological support on activities related to their standardisation.

The rollout of 5G/6G networks and large-scale deployments of cellular IoT will lead to fundamental changes to our society, impacting not only consumer service but also industries embarking on digital transformations. While their standardisation and definition processes are ongoing, the key challenges are the lack of accurate, fast, low-cost, and traceable methods for the verification of new radio (NR) high-volume products and this is mainly due to lack of adoption of data sceince and machine learning in such field where automation can be applied through data-centric processes and engineering that will be able to adapt to changing environments and requirements.

The main objective will be to conduct extensinve data gathering on inteliigent antennas, radio channels, over the air test beds and finally create accurate and adaptive radio link models informed by traceable and data-centric learning algorithms. 

Supervisor: Dr Akram Alomainy

Smart data approach to improved product safety assessment

Every year there are products available in the UK market that pose a serious risk to the health and safety of consumers. Product risk assessment is the overall process of determining whether a product, which could be anything from a washing machine to a teddy bear, is judged safe for consumers to use.

There are several methods of product risk assessment, including RAPEX, which is the primary method used by product safety regulators in the UK and EU. However, despite its widespread use, the Office of Product Safety and Standards (OPSS) – the UK product safety regulator – has identified several limitations of RAPEX that it feels could be addressed by exploiting state-of-the-art ‘smart data’ methods. In this project OPSS will provide us with in kind contributions including at least access to data sets, and access to policy agendas that will help address the identified limitations of RAPEX. These include societal/population risk, hazard exposure and risk tolerability since RAPEX does not consider these critical factors when estimating product risk.

It is proposed to use Bayesian Networks (combining data and causal knowledge) to produce a new method of product risk assessment to address these factors. This project will combine expertise in computer science, statistics, psychology, and risk assessment at QMUL (Risk and Information Management Group) and OPSS and use different research methods such as case studies. It will contribute to the limited literature on quantitative product risk assessment and provide a causal systematic method and tool for product risk assessment that effectively estimate and communicate product risk and uncertainty. Finally, since this project is built around the needs of OPSS, we believe that its output will directly impact and improve the risk assessment process for regulators and manufacturers ensuring that the products we use are acceptably safe.

Supervisor: Prof Norman Fenton

Parametric controls from data analytics

Many production technologies have been developed with experts in mind. Thus, their controls require specialist knowledge. Yet, it is now possible to get a wealth of information on how a technology is used to achieve a task, and to gather descriptions of their use (‘make the image brighter’, ‘show me just where to place the fixtures’, ‘remove the excess noise’,…). This information may be harnessed to map inputs and control settings to meaningful targets. Such data analytics allows the construction of high level, parametric controls which make the tools more accessible and more intuitive. The goal of this research project is to explore, implement and evaluate different approaches to this challenge. Though the focus will be on real world data sets and practical applications, it will lead to core results that can generalise to different data sets and domains.

Supervisor: Prof Josh Reiss

Deep audio inpainting 

Real-life audio signals often suffer from local degradation and lost information. Examples include short audio intervals corrupted by impulse noise and clicks, or a clip of audio wiped out due to damaged digital media or packet loss in audio transmission. Audio inpainting is a class of techniques that aim to restore the lost information with newly generated samples without introducing audible artifacts. In addition to digital restoration, audio inpainting also finds wide applications in audio editing (e.g. removing audience noise in live music recording) and music enhancement (e.g. audio bandwidth extension and super-resolution). Approaches to audio inpainting can be classified depending on the length of the lost information, i.e. the gap. For example, in declicking and declipping, corruption may be frequently but mostly confined to only a few milliseconds duration or less. On the other hand, gaps on a scale of hundreds of milliseconds or even seconds may happen due to digital media damage, transmission loss, and audio editing. While intensive work has been done on inpainting short gaps, long audio inpainting still remains a challenging problem due to the high dimensional, complex and non-correlated audio features.

This project intends to investigate the possibility of adapting deep learning frameworks from various domains inclusive of audio synthesis and image inpainting for audio inpainting. A particular focus will be given to recovering musical signals with long-gap information missing, and reconstructing super-resolution audio signals through bandwidth extension, which are both challenging tasks in the state of the art. The research will be conducted by combining one or several methodologies from:
1) Traditional musical signal processing approaches, e.g. exemplar-based method.
2) Deep learning based approaches, e.g. convolution neural networks and generative adversarial network.
3) Audio-visual based approaches exploiting additional visual context, e.g. video recording of instrument performance.

Supervisor: Dr Lin Wang

Smarter Mobility Profiling Towards Enabling Low Carbon Footprint Lifestyles and Businesses

Citizens, transport and building providers, transport face key gaps in knowledge or challenges to accurately profile end-to-end journeys. The potential impact of this project is huge to provide a more complete insight into a more complete activity profiler to actionably shift behaviour to promote low-, or zero- carbon travel, and safer, healthier passenger travel. The main aim of this project is to develop a smarter than state of the art profiler for activities of daily human mobility including those occuring indoors. The specific sub-objectives are to investigate how to: select which optimal combination of sensors can provide a more complete carbon footprint for end to end urban travel and/or building usage; deal with sparse data sets for specific motion activities. The methodology would seek to extend existing work: to better estimate outdoor, combined with indoor, travel trajectories; it would handle noisy, incomplete, sensing; investigate algorithms based upon dynamic time warping, extended Kalman & particle filters, graph theory and low-shot machine learning.

Supervisor: Dr Stefan Poslad