In this Turing Artificial Intelligence Acceleration Fellowship, I will focus on artificial intelligence for medical treatments and therapies. I take the view that AI is a question on how to realise artificial systems that solve practical problems currently requiring human intelligence to solve, such as those solved by clinicians, nurses and therapists. Critical care is high risk and highly invasive environment caring for the sickest patients at greatest risk of death. Patients within this environment are highly monitored, enabling sudden changes in physiology to be attended to immediately. In addition, this monitoring requires a heavier staffing ratio (often 1:1 nursing; 1:8 medical) and variances in human factors and non-technical pressures (e.g. staffing, skill-mix, finances) leads to critical care delivery being disparate. AI in healthcare is a hard problem as, due to the diversity and variability of human nature, systems have to cope with unexpected circumstances when solving perceptual, reasoning or planning problems. Crucially, AI has two facets: Understanding from data, and Agency. While rapid strides have been made on learning from data, e.g. how to make medical diagnosis more precise and faster than human experts, there is little work on how to carry on after the diagnosis, e.g. which therapy and treatment to conduct. The latter requires agency and has seen fewer applications as it is a harder problem to solve. My clinical partners and I want to develop the required AI algorithms that can learn and distil the best plan of action to treat a specific patient, from the expert knowledge of clinicians. We will focus on an area of AI called RL that has been successful in enabling robots and self-driving cars to learn a form of autonomous agency. We want to transform these methods into the healthcare domain. This will require the development of new RL algorithms, able to efficiently understand the state of a patient from noisy and ambiguous hospital data. The system will not only learn to recommend interventions such as prescribing drugs and changing dosages as needed per patient but to make these recommendations in a manner that is meaningful to the clinical decision-makers and helps them make the best final decision on a course of action. The methods developed as part of this project can be used in different applications beyond healthcare. Many sectors within industry, such as aerospace, or energy, deal with similar bottlenecks. These are highly regulated environments, with great need for decisions making support, but a scarcity of highly skilled human experts. With sufficient data, our methods can be applied to these sectors as well, to distil the required human expertise and best practices from top experts, and use them to drive decision making all over the sector, for increased efficiency and safety.

This proposal is for a new EPSRC Centre for Doctoral Training in Wind and Marine Energy Systems and Structures (CDT-WAMSS) which joins together two successful EPSRC CDTs, their industrial partners and strong track records of training more than 130 researchers to date in offshore renewable energy (ORE). The new CDT will create a comprehensive, world-leading centre covering all aspects of wind and marine renewable energy, both above and below the water. It will produce highly skilled industry-ready engineers with multidisciplinary expertise, deep specialist knowledge and a broad understanding of pertinent whole-energy systems. Our graduates will be future leaders in industry and academia world-wide, driving development of the ORE sector, helping to deliver the Government's carbon reduction targets for 2050 and ensuring that the UK remains at the forefront of this vitally important sector. In order to prepare students for the sector in which they will work, CDT-WAMSS will look to the future and focus on areas that will be relevant from 2023 onwards, which are not necessarily the issues of the past and present. For this reason, the scope of CDT-WAMSS will, in addition to in-stilling a solid understanding of wind and marine energy technologies and engineering, have a particular emphasis on: safety and safe systems, emerging advanced power and control technologies, floating substructures, novel foundation and anchoring systems, materials and structural integrity, remote monitoring and inspection including autonomous intervention, all within a cost competitive and environmentally sensitive context. The proposed new EPSRC CDT in Wind and Marine Energy Systems and Structures will provide an unrivalled Offshore Renewable Energy training environment supporting 70 students over five cohorts on a four-year doctorate, with a critical mass of over 100 academic supervisors of internationally recognised research excellence in ORE. The distinct and flexible cohort approach to training, with professional engineering peer-to-peer learning both within and across cohorts, will provide students with opportunities to benefit from such support throughout their doctorate, not just in the first year. An exceptionally strong industrial participation through funding a large number of studentships and provision of advice and contributions to the training programme will ensure that the training and research is relevant and will have a direct impact on the delivery of the UK's carbon reduction targets, allowing the country to retain its world-leading position in this enormously exciting and important sector.

High Performance Embedded and Distributed Systems (HiPEDS), ranging from implantable smart sensors to secure cloud service providers, offer exciting benefits to society and great opportunities for wealth creation. Although currently UK is the world leader for many technologies underpinning such systems, there is a major threat which comes from the need not only to develop good solutions for sharply focused problems, but also to embed such solutions into complex systems with many diverse aspects, such as power minimisation, performance optimisation, digital and analogue circuitry, security, dependability, analysis and verification. The narrow focus of conventional UK PhD programmes cannot bridge the skills gap that would address this threat to the UK's leadership of HiPEDS. The proposed Centre for Doctoral Training (CDT) aims to train a new generation of leaders with a systems perspective who can transform research and industry involving HiPEDS. The CDT provides a structured and vibrant training programme to train PhD students to gain expertise in a broad range of system issues, to integrate and innovate across multiple layers of the system development stack, to maximise the impact of their work, and to acquire creativity, communication, and entrepreneurial skills. The taught programme comprises a series of modules that combine technical training with group projects addressing team skills and system integration issues. Additional courses and events are designed to cover students' personal development and career needs. Such a comprehensive programme is based on aligning the research-oriented elements of the training programme, an industrial internship, and rigorous doctoral research. Our focus in this CDT is on applying two cross-layer research themes: design and optimisation, and analysis and verification, to three key application areas: healthcare systems, smart cities, and the information society. Healthcare systems cover implantable and wearable sensors and their operation as an on-body system, interactions with hospital and primary care systems and medical personnel, and medical imaging and robotic surgery systems. Smart cities cover infrastructure monitoring and actuation components, including smart utilities and smart grid at unprecedented scales. Information society covers technologies for extracting, processing and distributing information for societal benefits; they include many-core and reconfigurable systems targeting a wide range of applications, from vision-based domestic appliances to public and private cloud systems for finance, social networking, and various web services. Graduates from this CDT will be aware of the challenges faced by industry and their impact. Through their broad and deep training, they will be able to address the disconnect between research prototypes and production environments, evaluate research results in realistic situations, assess design tradeoffs based on both practical constraints and theoretical models, and provide rapid translation of promising ideas into production environments. They will have the appropriate systems perspective as well as the vision and skills to become leaders in their field, capable of world-class research and its exploitation to become a global commercial success.

Enterprises and government entities have a growing need for systems that provide decision support based on descriptive and predictive analytics over large volumes of data. Examples include supporting decisions on pricing and promotions based on analyses of revenue and demand data; supporting decisions on the operation of complex equipment based on analyses of sensor data; and supporting decisions on website content based on analyses of user behaviour. Such support may be critical for safety and regulatory compliance as well as for competitiveness. Current data analytics technology and workflows are well-suited to settings where the data has a uniform structure and is easy to access. Problems can arise, however, when performing data analytics in real-world settings, where as well as being large, datasources are often distributed, heterogeneous, and dynamic. Consider, for example, the case of Siemens Energy Services, which runs over 50 service centres, each of which provides remote monitoring and diagnostics for thousands of gas/steam turbines and ancillary equipment located in hundreds of power plants. Effective monitoring and diagnosis is essential for maintaining high availability of equipment and avoiding costly failures. A typical descriptive analytics procedure might be: "based on sensor data from an SGT-400 gas turbine, detect abnormal vibration patterns during the period prior to the shutdown and compare them with data on similar patterns in similar turbines over the last 5 years". Such diagnostic tasks employ sophisticated data analytics tools, and operate on many TBs of current and historical data. In order to perform the analysis it is first necessary to identify, acquire and transform the relevant data. This data may be stored on-site (at a power-plant), at the local service centre or at other service centres; it comes in a wide range of different formats, ranging from flat files to XML and relational stores; access may be via a range of different interfaces, and incur a range of different costs; and it is constantly being augmented, with new data arriving at a rate of more than 30 GB per centre per day. Acquiring the relevant data is thus very challenging, and is typically achieved via a combination of complex queries and bespoke data processing code, with numerous variants being required in order to deal with distribution and heterogeneity of the data. Given the large number of different analytics tasks that service centres need to perform, the development and maintenance of such procedures becomes a critical bottleneck. In ED3 we will address this problem by developing an abstraction layer that mediates between analytics tools and datasources. This abstraction layer will adapt Ontology Based Data Access (OBDA) techniques, using an ontology to provide a uniform conceptual schema, declarative mappings to establish connections between ontological terms and data sources, and logic-based rewriting techniques to transform ontological queries into queries over the data sources. For OBDA to be effective in this new setting, however, it will need to be extended in several different directions. Firstly, it needs to provide greatly extended support for basic arithmetic and aggregation operations. Secondly, it needs to deal more effectively with heterogeneous and distributed data sources. Thirdly, it will be necessary to support the development, maintenance and evolution of suitable ontologies and mappings. In ED3 we will address all of these issues, laying the foundations for a new generation of data access middleware with the conceptual modelling, query processing, and rapid-development infrastructure necessary to support analytic tasks. Moreover, we will develop a prototypical implementation of a suitable abstraction layer, and will evaluate our prototype in real-life deployments with our industrial partners.

One of the main contributors towards the cost of high-value engineering assets is the cost of maintenance. Taking an aircraft out of service for inspection means loss of revenue. However, if damage occurs and leads to catastrophic failure, safety and casualties are major issues. In terms of an offshore wind farm, the cost of an unscheduled visit to a remote ocean site to replace a 75m blade is exceedingly high. If one adopts an approach to maintenance where the structure of interest is monitored constantly by permanent sensors, and data processing algorithms alert the owner or user when damage is developing, one can optimise the maintenance programme for cost without sacrificing safety. If damage is detected early, repair rather than replacement can be viable. The complexity of modern structures and their challenging operating environments make it difficult to develop algorithms that can detect and identify early damage. The relevant discipline - structural health monitoring (SHM) - suffers from problems that have prevented uptake of the technology by industry. Although structural complexity makes analysis difficult, one variant of SHM - the data-based approach - shows great promise. In this case one uses machine learning techniques to diagnose damage from measured data. Data-based SHM faces a number of challenges; the first is that most data-based approaches to SHM require measured data from the structure in all possible states of damage. For a structure like an 5 MW wind turbine - it is simply not conceivable that one should damage a single one for data collection purposes, let alone many. Fortunately, if one is only interested in whether damage is present or not, this is possible using only data from the healthy condition. One builds a picture of the healthy state of the structure and then monitors for deviations. This raises a second issue with data-based SHM; if one is monitoring the structure for changes, one does not wish to be deceived by a benign change in its environmental/operational conditions - so-called 'confounding influences'. The original Fellowship aimed to solve these problems via a population-based approach to SHM modelled on the discipline of 'syndromic surveillance' (SS), which is used to detect disease outbreaks in human populations. The core of the proposed research was an intelligent database holding data across populations of structures, and an inference engine that could use damage data from an individual, to allow diagnostics on others. The original work has progressed very well; the required database was created and algorithms for inference across populations have been developed and demonstrated. Algorithms for removing confounding influences have also been created which are arguably now the state of the art. The Fellowship so far has also allowed insights into how population-based SHM can go far beyond technologies based on SS, leading to this new proposal. Very new concepts in SHM will be explored. The first idea is to extend the 'database' to an 'ontology'; ontologies encode, share and re-use domain knowledge. In a way, moving to an ontology adds a 'language centre' to the existing storage and processing; one might even think of the result as a computational brain concentrating on a specific engineering field - in this case SHM. New population-based methods are proposed. For populations of near-identical structures, the idea of the 'form' of a structure is presented. The form is created to represent all individuals in a population, if damage data are available for an individual turbine in a wind farm, they can be transferred into the form and thus allow inference across the farm. Furthermore, a general theory of populations of disparate structures will be constructed using ideas from mathematics and computation: geometry, graph theory, complex networks and machine learning. Again, the theory will allow damage data from individuals to generate insights across the population.