AMBS has begun a collaboration with major UK defence company BAE Systems.
The four-year project, worth around £130,000, is funded by both BAE Systems and the Engineering and Physical Sciences Research Council (EPSRC), and starts this year.
It will look at the use of more automated decision support tools in the defence industry, and especially at the use of Reinforcement Learning (RL) algorithms within military scenarios. In particular, BAE Systems is currently working up concepts for the next generation of naval vessels and developing semi-automated intelligent decision aids which are being incorporated into its INTeACT combat management system product.
As part of the process the company is keen to tap into the knowledge of data scientists who can help it integrate decision sciences methodologies into its software in the safest way possible.
Dr Richard Allmendinger, Business Engagement Lead at AMBS and Senior Lecturer in Data Science, said the results from the project will help develop further confidence in the deployment of RL algorithms in the military field, and especially in operational scenarios where environmental conditions inevitably differ from simulated environments.
"The project is all about understanding how defence systems software can deal with uncertainty better and draw on prior experiences to make informed decisions. It is also about the industry itself learning further about developing the necessary governance processes associated with the increased use of machine learning technologies.
"What BAE Systems is particularly interested in from us is a further understanding of the methodologies involved, because in this area the technologies are so complex that no one person is capable of digesting all the information available and fully understanding it. But this is not about replacing people, it is about making decision support tools work better."
The project is all about understanding how defence systems software can deal with uncertainty better.
The study will conduct research into the trade-off between the generality of RL training environments and real-world performance. It will also look at how the performance and robustness of RL algorithms can be characterised within a changing environment, and whether the performance drop-off thresholds can be measured and predicted.
Dr Allmendinger said the ICASE scheme was an excellent platform because there were clear benefits for both the company and the university partner. “ICASE studentships not only provide an opportunity for PhD researchers to gain first-hand experience of working outside an academic environment, but the student is also analysing real-time, real-life projects which could potentially lead to wider commercialisation, so there is really strong impact. Such schemes can also typically lead to additional projects between the university concerned and the company."
Simon Mettrick from BAE Systems said: “The current ICASE PhDs at The University of Manchester have been hugely helpful in deepening understanding of trust in artificial intelligence and potential applications of computer vision. This opportunity promises to do the same for reinforcement learning."
Louise Bates, Head of Strategic Partnerships at The University of Manchester, said: “This is another great example of The University of Manchester partnering with BAE Systems to address a real world challenge, and it further demonstrates the value of our strategic partnership."
Thu Trang Dinh, AMBS PhD Researcher
AMBS PhD researcher Thu Trang Dinh has been recognised by one of the most prestigious organisations in data science. She attended a Data Study group organised by The Alan Turing Institute in London, and at the end of an intensive three-week exercise was awarded a Learning Machine Award which recognises individuals who absorb new skills, tackle unfamiliar methods, and are not afraid to try new techniques.
The study groups attract students from across the UK and are run as intensive ‘collaborative hackathons’ where researchers are set specific challenges and work together on real-world problems.
Her group was set a challenge by the UK Dementia Research Institute which wants to generate predictions for the effects of Alzheimer’s associated genetic variants on cell-type specific gene regulatory mechanisms.
Her own PhD project at AMBS is looking at how machine learning methods can help predict exchange rate movements. As she explained: “Predicting and forecasting accurate exchange rates can bring a host of benefits for policymakers and any business that trades abroad, and I am currently working on a machine learning predictive model that incorporates many different sources of information. Banks and financial institutions are today starting to use machine learning much more widely and this is just one example of the benefits it can bring.”
Andreea Avramescu PhD Researcher
Andreea Avramescu, a PhD researcher in Decision Sciences at AMBS, has been awarded a prestigious SPECIES Scholarship award. The scholarship aims to promote evolutionary algorithmic thinking and winners spend three months at a partner institution carrying out research. Andreea has chosen to undertake her research placement at the University of Malaga and her project examines the complex optimisation problems involved in the optimal design of supply chains.
As she explains: “I am particularly interested in optimisation problems addressing social sustainability issues in the fields of logistics, supply chain management, and applied optimisation. I have previously worked on topics related to international migration, social media data, sentiment analysis, cryptomarkets, drug consumption, and illegal supply chains in both industry and academia."
Her PhD research, under the supervision of Dr. Richard Allmendinger at AMBS, draws on tools from operations research and machine learning to guide the optimisation of the manufacturing tasks and delivery strategies of personalised medical products.
Dr. Darminder Singh Ghataoura, Director of AI & Data Science, Fujitsu UK
A data science expert has been awarded a Simon Industrial and Professional Fellowship at AMBS to research on the role of trust in AI systems. Dr. Darminder Singh Ghataoura is Director of AI & Data Science within the Defence & National Security division of Fujitsu UK, and was brought in by the company to build and lead its offerings and capabilities in AI, data science and data strategy.
He has now started a six-month Fellowship looking at the important areas of trustworthy AI and AI decision making in the defence sector and elsewhere. For example he is looking at how simulation techniques and gaming technologies could be used by the defence industry to help address supply chain logistics in specific scenarios.
As he explained: “As the defence industry integrates AI into its systems and missions there are questions about the role of trust in human-machine teams. To address these concerns there have been continuing technological advancements that look to build trust into AI systems by enhancing system functions and features like transparency, explainability, auditability, reliability, robustness, and responsiveness.
“However little effort has been made to address solutions which exploit research on human attitudes towards ‘machine-teaming’, accounting for the differences in people’s perceptions and experiences, especially within dynamic and changing environments such as a battlefield scenario.”