1. Home
  2. The University
  3. Our Faculties
  4. Faculty of Science and Engineering
  5. School of Mathematics and Computer Science
  6. Computational Foundry
  7. EPSRC Centre for Doctoral Training
  8. Work on Difficult Problems in Challenging Contexts

Work on Difficult Problems in Challenging Contexts

Keyboard with tags
Black Friday
Cyber security, cursor on a monitor

Challenge Contexts

Our challenge contexts— from health and wellbeing; industry 4.0 and smart manufacturing; cyber threats, ethics and law; sustainable energy; and digital economy platforms, services and innovation —are ones that are emerging as key drivers of our region’s and nation’s industrial and societal strategies. PhD researchers will work together across cohorts to interrogate their scientific methods and insights through multiple contexts and scenarios; an approach essential in ensuring complex, varied needs and values are accommodated.

We will meet the need to train researchers who can work to ensure that such systems are trustable, understandable, usable and negotiable by and with humans. Without such capability, the hoped-for benefits of these key technologies will be severely undermined. We subscribe to and will work out a computational response to the urgent need articulated in the Royal Society and British Academy principle of “human flourishing” and the problems that will occur if this is not adopted.

Our people-first perspective will drive the invention and understandings of new theories, algorithms, data, platforms and interaction devices. It will question the orthodoxy of well-established approaches by considering them from the perspective of the people, organisations and societies they are meant to serve. Our approach goes far beyond the application of systems to real-world problems; rather, it takes real-world contexts and uses them to re-form, shape and replace the computational science needed to make such systems useful, usable, satisfying, delightful and acceptable by society.

You can find out more about each of our current PhD Researchers projects and Challenging contexts here.

 

Sample problems

Flexible collaboration system architectures for human-robot interactions on the assembly line. Automated robots play an important role in modern manufacturing, especially on the assembly line. The increasing pace of introduction of new products, particularly those with many variants and short lifetimes, brings considerable uncertainty to assembly line design; further, in order to reduce initial investment, manual assembly may be more preferable. It is therefore very desirable to be able to develop flexible and reconfigurable manufacturing systems. PhD researchers will investigate solutions to exploit human-robot physical cooperation considering how human factors and safety issues along with workers’ insights into efficient and fulfilling work can lead to both novel multimodal interaction techniques and agily adaptable system architectures to close the gap between fully manual assembly and fully automated manufacturing lines.

Mixed intelligences: Integrating crowdsourcing and AI approaches. In a pilot project, we have deployed a human-powered smart speaker system in the streets of Dharavi (a vast slum area in Mumbai). At the same time, we have installed standard AI-powered Google “Home” devices. We will use the data from both deployments to consider how a mixed intelligence approach can provide better, more human-like, answers. This architecture can be applied to many other contexts in their aspirations to mix local community perspectives on places with their vast and diverse sensor-driven and other data sets.

Innovating system configurations and interfaces through exposure to understandings of terrorist groups’ exploitation of data-driven and intelligent technologies. With society’s increasing dependence on these systems, they will become greater targets for bad actors and these groups will also use their own systems for negative impacts. Using insights directed by our cyber threats teams on the mode of operation and motivations of such groups, PhD researchers will uncover novel verification techniques that might identify any tampering of critical systems.

Sample problems

Novel image processing to meet health professional needs by creating a virtual pathology lab. Health professionals are highly skilled in considering diverse and complex patient test results. By considering their processes and practices, PhD projects can lead to automation pipelines that better support these abilities. For example, millions of blood tests are carried out every year in the UK. The processes used are usually highly automated but the nationwide throughput is limited. Furthermore, many exams cannot be carried out in parallel, so the more prescribed, the more expensive the bill is. A disruptive approach would be to consider the feasibility of a virtual pathology lab where samples would be acquired from either 3D microscopy or nano-CT devices. Ideally, a virtual pathology lab would scan one human sample and create as many exam outputs as possible at once as the cost of all these tests shall be no more than the cost of a single test. This would allow health professionals to have a wide overview of patient health and lead to faster diagnoses.

Deep learning to assist in complex decision making for societal and economic benefits. The ability to explain the rationale behind a decision is important in many applications, such as in financial transaction modelling to understand potential investment opportunities or medical diagnosis where clinicians would like to verify and interpret the decision made by machine-learning models. Whilst data-driven modelling has been shown to be capable to produce human-level performance, the decision making of these complex models is opaque and its inference process differs significantly from human processing. Users, therefore, have difficulties to understand or trust their outcomes. The aim is to make deep learning processes humanly interpretable by introducing new methods and integrating visualisation techniques that convert this “black-box” deep learning to “glass-box”/“white-box" computing.