The focus of this proposal is on the detection and survival of wrong code compiler defects, which we argue present a cyber-security threat that has been largely ignored to date. First, incorrectly compiled code can introduce exploitable vulnerabilities that are not visible at the source code level, and thus cannot be detected by source-level static analysers. Second, incorrectly compiled code can undermine the reliability of the application, which can have dramatic repercussions in the context of safety-critical systems. Third, wrong code compiler defects can also be the target of some of the most insidious security attacks. A crafty attacker posing as an open source developer can introduce a compiler-bug-based backdoor into a security-critical application by adding a patch that looks perfectly innocent but which, when compiled with a certain compiler, yields binary code that allows the attacker to compromise the software. In this project, we aim to explore automated techniques that can detect and prevent such problems. In particular, we plan to investigate techniques for automatically finding compiler-induced vulnerabilities in real software, approaches for understanding the extent to which an attacker could maliciously modify an application to create a compiler-induced vulnerability, and methods for preventing against such vulnerabilities at runtime.

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.

Systems modelled by partial differential equations (PDEs) are ubiquitous in science and engineering. They are used to model problems including structures, fluids, materials, electromagnetics, wave propagation and biological systems, and in areas as varied as aerospace, image processing, medical therapeutics and economics. PDEs comprise a forward model for predicting the response of a system, but are also a key component in the solution of inverse problems, for design optimisation, uncertainty quantification and data science applications, where the forward computation is repeated many times with different inputs. The numerical simulation of complex systems modeled by PDEs is a challenging topic. It involves the choice of underlying equations, the selection of suitable numerical solvers, and implementation on specific hardware. Over the decades numerous software libraries have been developed to support this task. But adapting these libraries to the specific model and combining the various components in a low-level high-performance programming language requires a major development effort. This required effort has become significantly more challenging with the advent of heterogeneous mixed CPU/GPU devices on the path to exascale systems. Implementations need to be adapted for each individual device type in order to achieve good performance. As a consequence, developing new simulations at scale has become an ever more costly and time-intensive task. In this project we propose a different simulation paradigm, based on the use of high-productivity languages such as Python to describe the problem, and automatic code generation and just-in-time compilation to translate the high-level formulations into high-performance exascale-ready code. Based on the experience with the component software libraries Firedrake, FEniCS and Bempp, the investigators will build a toolchain for complex exascale simulations of PDEs on unstructured grids, using state of the art finite element and boundary element technologies. The research will include mathematical and algorithmic underpinnings, concrete software development for automatic code generation of low-level CPU/GPU kernels, high-productivity language interfaces, and the application to 21st century exascale challenge problems in the areas of battery storage systems, net-zero flight, and high-frequency wave propagation.

Machine Learning (ML) already has a dramatic impact on our daily lives. ML developments in large language models and deep generative models cement that further. The recent explosion in ML, however, is built on the back of improved computer systems able to train and generate ever more powerful models. Systems design fundamentally defines ML performance and capability. This is true for Internet-scale ML and artificial intelligence (AI). Yet, more recently, it is especially evident in distributed, efficient, device-oriented, secure, personalised, privacy-preserving ML. UK strength in this fast developing area is dependent on a skilled R\&D workforce. Systems research and ML research are symbiotic. Current innovation in systems research is driven by the ubiquitous need for efficient and reliable ML. ML research, conversely, is steered by deployment capability and the economic and environmental impact of the resulting systems. Furthermore, systems research increasingly relies on ML methods to automate design, and ML research develops such methods. Major gains are made when the development of ML and systems are co-developed and co-optimized. This is relevant across a broad spectrum of industries: in-car systems, medical devices, mobile phones, sensor networks, condition monitoring systems, high-performance compute and high-frequency trading. Yet PhD training that brings together systems and ML is rare; research training is often siloed in the individual sub-disciplines. Instead, we need researchers trained in both fields and experienced in working across them. Hence: The ML Systems CDT will train a new type of student -- the ML-systems researcher. The ML Systems researcher is critically capable in both fields, and has collaborative research experience across the systems-ML stack. An example concretises this. A company is developing and deploying wearable body monitors. Effective models must be learnt on collected data, but data must be privacy preserving and bandwidth minimized. This is then personalised to each individual, adaptable to circumstance while being battery efficient and not connection dependent. To manage such a project requires knowledge of effective data-efficient ML signal analysis methods, designed and optimized for low-power hardware, itself tailored for the purpose through ML optimization methods. Knowledge of personalisation methods and the payoffs of privacy preserving methods vitally complement this. The societal impact, e.g.\ on those who might be obsessive about their medical state must also be considered, and will impact development. This CDT will train individuals with cross-cutting capability in all these components. Students must have broad understanding of different hardware designs, different platforms, different environments, different models, and different goals beyond their immediate research focus. This makes a cohort-based CDT vital. Standard PhD training in ML systems can result in research focus on a single ML technique and a single system. The CDT treats ML Systems as a holistic discipline. Cohort interaction, and integration gives students real experience across multiple systems, approaches and methodologies. Furthermore students will join together to contribute to a unified toolkit for the ML-Systems stack, and make use of others' contributions to that toolkit. On leaving the CDT, our graduates will understand fully where to focus resources to best improve a company's real-world ML development - whether that be at the ML-algorithm level, the hardware level, the compiler, level or even the legal level. They will be able to evaluate work at every level. We expect our graduates to be the leading team managers in real-world cutting-edge company ML.

Users want mobile devices that appear fast and responsive, but at the same time have long lasting batteries and do not overheat. Achieving both of these at once is difficult. The workloads employed to evaluate mobile optimisations are rarely representative of real mobile applications and are oblivious to user perception, focussing only on performance. As a result hardware and software designers' decisions do not respect the user's Quality of Experience (QoE). The device either runs faster than necessary for optimal QoE, wasting energy, or the device runs too slowly, spoiling QoE. SUMMER will develop the first framework to record, replay, and analyse mobile workloads that represent and measure real user experience. Our work will expose for the first time the real Pareto trade-off between the user's QoE and energy consumption. The results of this project will permit others, from computer architects up to library developers, to make their design decisions with QoE as their optimisation target. To show the power of this new approach, we will design the first energy efficient operating system scheduler for heterogeneous mobile processors which takes QoE into account. With heterogeneous mobile processors just now entering the market, a scheduler able to use them optimally is urgently needed. We expect our scheduler to be at least 50% more energy efficient on average than the standard Linux scheduler on an ARM BIG.LITTLE system.