Department of Computer Science, University of Oxford Projects All

Department of Computer Science, University of Oxford Projects All http://www.cs.ox.ac.uk/projects/ All en-gb 360 Wed, 13 May 2026 19:06:59 GMT Wed, 13 May 2026 19:06:59 GMT All http://blogs.law.harvard.edu/tech/rss/ Algorithmic Proofs of Algorithmic Impossibility http://www.cs.ox.ac.uk/projects/AlgorithmicProofsofAlgorithmicImpossibility/ The project will exploit and strengthen synergies between algorithms and lower bounds in order to make progress in major questions in circuit complexity and the theories of cryptography and learning. The main objectives are the following: I. Frontier circuit lower bounds: To develop methods based on combinatorics and discrete analysis to solve longstanding open questions in circuit complexity, making progress towards showing that P is not equal to NP. II. Lower bounds for algorithms: To strengthen the algorithmic consequences of proofs of lower bounds, applying modern combinatorial constructions. Besides new learning algorithms, this direction can refute heuristic cryptographic constructions which were recently proposed. III. A theory of quantum meta-complexity: To develop the above synergies in the context of quantum computing, a very timely investigation in view of the rise of quantum cryptography. This includes circuit lower bounds from quantum algorithms, and connections between quantum learning and quantum cryptography. Department of Computer Science, University of Oxford Projects All oucl-All-Algorithmic Proofs of Algorithmic Impossibility Wed, 13 May 2026 19:06:59 GMT ScanSpot: Intelligent Monitoring of Energy and Logistics Infrastructure with Multi-Modal Sensing http://www.cs.ox.ac.uk/projects/ScanSpot/ ScanSpot is an Innovate UK-funded research project developing a next-generation monitoring platform for energy and logistics infrastructure. It addresses growing challenges in site safety, security, and situational awareness, particularly in complex environments such as EV charging hubs and freight depots where existing CCTV-based systems are often unreliable. The system combines multi-modal sensor fusion, including LiDAR, radar, RGB-D, thermal, and low-light imaging, with deep learning to create a real-time 4D digital twin of a site. By continuously learning normal patterns of activity, ScanSpot can automatically detect anomalies such as unauthorised access, unusual vehicle behaviour, and early-stage thermal risks, enabling faster and more reliable incident detection. In collaboration with industry partners including Aegis Energy, the project aims to deliver scalable intelligent infrastructure monitoring that supports safer and more efficient operations while reducing risk for operators and insurers. Ultimately, ScanSpot contributes to the development of resilient and intelligent systems that underpin the transition to sustainable energy and transport networks. Department of Computer Science, University of Oxford Projects All oucl-All-ScanSpot Wed, 13 May 2026 19:06:59 GMT Meta-complexity: A Unified Approach to the Complexity of Proofs and Computation http://www.cs.ox.ac.uk/projects/Meta-complexity/ One of the most fundamental questions in computer science is the P vs NP question, which asks if every computational problem with efficiently verifiable solutions is efficiently solvable. Equivalently, it asks if all propositional tautologies have short proofs that can be found efficiently. The answer is widely believed to be negative, but we lack a rigorous justification for this belief. The field of computational complexity approaches P vs NP and related questions by showing lower bounds (i.e., impossibility results) on efficient computations, while the field of proof complexity approaches these questions by showing lower bounds on efficient proofs for propositional tautologies. Despite much effort, the best-known lower bounds in both computational complexity and proof complexity are quite far from resolving the P vs NP question, and there are significant barriers to the success of known techniques. In this project, we will approach fundamental lower bound questions in computational complexity and proof complexity using the novel conceptual framework of ``meta-complexity". Meta-complexity studies the complexity of computational problems and propositional statements that are themselves about complexity, e.g. the Minimum Circuit Size Problem, which asks if a given Boolean function has small Boolean circuits. Concepts and techniques from meta-complexity have been instrumental in major recent advances in theoretical cryptography and average-case complexity, overcoming known barriers. We will extend the methodology to attack some of the deepest questions in theoretical computer science, by showing new lower bounds on both proofs and computation, establishing strong connections between computational complexity and proof complexity, and giving applications to explicit constructions, learning and hardness of approximation. A key aspect of our approach is that meta-complexity is a unifying framework, which applies equally well to proofs and computation.   Department of Computer Science, University of Oxford Projects All oucl-All-Meta-complexity Wed, 13 May 2026 19:06:59 GMT Algorithmic Comparison of Stochastic Systems http://www.cs.ox.ac.uk/projects/AlgorithmicComparisonofStochasticSystems/ Computer systems are woven into almost every aspect of modern life, from aircraft and medical devices to the software that underpins transport, communication, and industry. As these systems grow ever more complex, their correctness and performance can no longer be assessed reliably by hand. Formal verification addresses this challenge by using mathematical and algorithmic methods to analyse systems automatically. This project focuses on the formal verification of probabilistic systems, whose behaviour involves randomness, whether through randomised algorithms or through uncertain and stochastic environments. The first aim of the project is to strengthen one of the central foundations of current probabilistic verification tools: probabilistic bisimilarity. This notion is crucial for combating the state-space explosion problem, since it allows verification tools to reduce large models by merging states that are behaviourally equivalent. Yet existing equivalence notions can be fragile: even very small changes in probabilities may produce disproportionately large changes in behaviour, potentially undermining the reliability of verification results, especially when parts of a system are only approximately known. The project therefore seeks robust new forms of probabilistic bisimilarity that preserve the practical benefits of minimisation while making verification results more stable and trustworthy. The second aim is to develop broader and more powerful methods for comparing probabilistic systems. This includes bringing ideas from dynamical systems theory, such as Lyapunov exponents, into algorithmic verification in order to better capture how behavioural differences evolve over time, with potential applications to diagnosability and privacy verification. The project will also design new algorithms that can handle much richer classes of systems, including those with infinitely many states, continuous observations, and more general forms of nondeterminism. In the longer term, this research aims to extend the foundations of probabilistic verification and enable a new generation of more expressive, reliable, and widely applicable verification tools. Department of Computer Science, University of Oxford Projects All oucl-All-Algorithmic Comparison of Stochastic Systems Wed, 13 May 2026 19:06:59 GMT CHAILD: Children’s Agency in the Age of AI: Leveraging Interdisciplinarity http://www.cs.ox.ac.uk/projects/CHAILD/ AI systems are increasingly central to children's digital environments, encompassing connected toys, apps, voice assistants, and online learning platforms. While these systems offer valuable opportunities for children’s development and learning, they also pose significant risks, such as screen time addiction and various forms of online harms. Current approaches to children's digital experiences are largely dominated by restrictive and protective measures, neglecting children’s potential to exercise control and make age-appropriate, informed choices themselves. While it may be challenging for younger children (especially those under five) to make meaningful decisions, such approaches make little contribution to fostering children’s autonomous agency in a digital context. There is an urgent need for new frameworks, guidelines, and strategies for policymakers, developers, parents, and educators to better support children’s development, well-being, and safety in an increasingly datafied environment. The two-year UKRI-funded project CHAILD—Children’s Agency in the Age of AI: Leveraging Interdisciplinarity—aims to establish a foundational understanding of children’s agency in the age of AI.  The project is led by the Department of Computer Science in collaboration with Institute for Ethics in AI and University College London (UCL Knowledge Lab).  Department of Computer Science, University of Oxford Projects All oucl-All-CHAILD Wed, 13 May 2026 19:06:59 GMT Challenges in Competitive Online Optimisation (CCOO) http://www.cs.ox.ac.uk/projects/CCOO/ Online decision-making, characterised by the need to make immediate decisions without knowledge of the future, lies at the heart of numerous applications. Yet our understanding of dealing with the resulting uncertainty remains poor. Through the lens of the established framework of online algorithms as well as the emerging field of learning-augmented algorithms, this project aims to address central challenges in decision-making under uncertainty. While there has been extensive research on online algorithms, many of the core challenges remain unresolved. However, several recent discoveries of novel algorithmic design and analysis techniques have opened up new avenues for overcoming previous obstacles. In parallel, the rise of machine learning is now dramatically enriching our tool-set for dealing with uncertainty. This has motivated the recent emergence of the field of learning-augmented algorithms. Here, an algorithm's input is augmented with predictions, aiming for near-optimal performance if predictions are reasonably good, while still retaining classical worst-case guarantees even for highly erroneous predictions. Inspired by these recent developments, this project aims to substantially elevate our understanding of decision-making under uncertainty. The main objectives are: <ol> <li>To explore new directions around the concept of work functions.</li> <li>To elevate the mirror descent technique to a generic tool for online algorithm design.</li> <li>To develop universal techniques for designing learning-augmented algorithms.</li> <li>To expand the scope of learning-augmented algorithms to new domains.</li> </ol> The project addresses questions at the forefront of theoretical computer science, building on Christian Coester's recent success in resolving several long-standing problems, and strives for foundational contributions to the timely issue of leveraging machine learning for improved algorithm design. Department of Computer Science, University of Oxford Projects All oucl-All-CCOO Wed, 13 May 2026 19:06:59 GMT Employing Categorical Probability Towards Safe AI http://www.cs.ox.ac.uk/projects/EmployingCategoricalProbabilityTowardsSafeAI/ This project is funded by the ARIA Safeguarded AI Programme. The last 5–10 years have seen the flourishing of categorical probability: category-theoretic techniques applied to probability theory. The power of category theory in this approach stems from its ability to organize and analyze the structure of statistical models and their composition. Categorical probability can be seen as a theoretical counterpart to the eminently successful method of probabilistic programming, which aims to structure and compose statistical models using methods from software engineering: this has already achieved practical success, through languages such as Stan and Pyro, across physical, biological and social sciences. In this first year of the project, we focus on bringing categorical probability to bear on three aspects that have been identified as crucial for world modelling in safe AI in the ARIA programme: • imprecise probability, giving bounds on probabilities of unsafe behaviour; • stochastic dynamical systems for world-modelling with random variables; • both with a solid underpinning of semantic version control. Department of Computer Science, University of Oxford Projects All oucl-All-Employing Categorical Probability Towards Safe AI Wed, 13 May 2026 19:06:59 GMT SCULI - Securing Convergent Ultra-large Scale Infrastructures http://www.cs.ox.ac.uk/projects/SCULI/ The SCULI (Securing Convergent Ultra-large Scale Infrastructures) programme, led by the University of Bristol, brings together experts from the Department of Computer Science, alongside counterparts at the Universities of Bristol and Lancaster, who will work across industry, policy and beyond, to forge a pioneering new approach to delivering cybersecurity.  The UK is the third most targeted country in the world for cyber-attacks, after the US and Ukraine, according to the House of Commons inquiry into Cyber Resilience of UK’s Critical National Infrastructure. From smart buildings, connected cities, smart farming and critical national infrastructure, the pace of change in digital technologies is increasing and so too is society’s dependence on them. As this increases, so does the risk of cyber-attacks and large-scale disruptions to essential infrastructure.  The SCULI programme draws on a unique mix of expertise - spanning sociotechnical approaches, and theoretical and applied computer science - and state-of-the-art lab facilities. It aims to transform the conceptualisation and delivery of cybersecurity in a world where connectivity has reached an unprecedented scale, with a prevalence of legacy and non-legacy systems, complex technology stacks and supply chains, and myriad intersections of humans and technologies.   Experts currently build and test cybersecurity approaches for components and systems at small scale and then attempt to upscale these to the infrastructures deployed to deliver services to society. The SCULI programme aims to transform this approach to securing such infrastructures, by instead understanding what the problems are at scale and designing solutions to work at that scale. The complexity and uncertainty cannot be removed. Instead, the programme embraces these issues as part of the problem and delivers solutions by both defining ideas and concepts, and designing and testing technical advances.   Key elements of the programme include:   <ul> <li data-leveltext="" data-font="Symbol" data-listid="3" data-list-defn-props="{"335552541":1,"335559685":720,"335559991":360,"469769226":"Symbol","469769242":[8226],"469777803":"left","469777804":"","469777815":"hybridMultilevel"}" aria-setsize="-1" data-aria-posinset="1" data-aria-level="1">A model that provides on-the-fly representation of cybersecurity goodness and new metrics to support cyber risk decision-making.  </li> <li data-leveltext="" data-font="Symbol" data-listid="3" data-list-defn-props="{"335552541":1,"335559685":720,"335559991":360,"469769226":"Symbol","469769242":[8226],"469777803":"left","469777804":"","469777815":"hybridMultilevel"}" aria-setsize="-1" data-aria-posinset="2" data-aria-level="1">New ways to compose and orchestrate security provision across variety of infrastructures with legacy and non-legacy elements. </li> <li data-leveltext="" data-font="Symbol" data-listid="3" data-list-defn-props="{"335552541":1,"335559685":720,"335559991":360,"469769226":"Symbol","469769242":[8226],"469777803":"left","469777804":"","469777815":"hybridMultilevel"}" aria-setsize="-1" data-aria-posinset="3" data-aria-level="1">Detection capabilities to assess with high accuracy, and at appropriate pace, the security state of such infrastructures throughout their operation to provide continuity of oversight and trust.</li> <li data-leveltext="" data-font="Symbol" data-listid="3" data-list-defn-props="{"335552541":1,"335559685":720,"335559991":360,"469769226":"Symbol","469769242":[8226],"469777803":"left","469777804":"","469777815":"hybridMultilevel"}" aria-setsize="-1" data-aria-posinset="4" data-aria-level="1">Incident response playbooks for such ultra-large-scale infrastructures and optimal ways to balance human-machine decision-making when infrastructures of such scale are under attack. </li> </ul> Department of Computer Science, University of Oxford Projects All oucl-All-SCULI Wed, 13 May 2026 19:06:59 GMT (De)constructing Quantum Software - DeQS http://www.cs.ox.ac.uk/projects/DeQS/ The past five years have seen quantum computation shift from a largely theoretical exercise to a practical one. Rather than imagining hypothetical quantum computers where we might one day run quantum algorithms, it is becoming increasingly possible to access and experiment with real, small-scale quantum hardware. These emerging devices show a great deal of potential, and may soon start to perform useful computations which are far beyond the reach of classical computers, solving hard problems in chemical development, material science, drug discovery, finance, industrial optimisation, or in areas that are, as yet, completely unexpected.   Just like classical computer hardware, quantum hardware is only as good as the software that drives it. This proposal focuses on quantum compilation, which concerns the translation of high-level descriptions of quantum algorithms to low-level instructions that can be run on specific quantum hardware. Current and near-future quantum hardware platforms are extremely limited in their capabilities and highly susceptible to noise from the environment, and sophisticated optimising compilers are crucial to practical applications for these devices.   The objective of this proposal is to produce more efficient and robust quantum software by means of advanced quantum compilation tools. In addition to that objective, the proposed research will develop new techniques to classically simulate large-scale quantum computations and prove they are correctly implemented. Whereas existing approaches treat the problems of compilation, classical simulation, and verifying correctness as separate challenges, the core idea behind this work is to treat these applications all as different aspects of the same thing: a flexible and generic notion of quantum program transformation based on graph rewriting. This provides a unique new perspective on these problems and allows a cross-pollination of ideas and techniques across these application areas, as well as between quantum computation and classical theoretical computer science.  Department of Computer Science, University of Oxford Projects All oucl-All-DeQS Wed, 13 May 2026 19:06:59 GMT WesCon – Observing the Evolving Structures of Turbulence (WOEST) http://www.cs.ox.ac.uk/projects/WOEST/ Our understanding of turbulent processes in the atmosphere is largely derived from theory and large eddy simulations (LES). There have been very few observations designed to evaluate the turbulent processes that are important for convection.   The Wessex Convection Experiment (WesCon) aims to address this gap using a range of ground-based instruments, the FAAM aircraft, and limited use of the Chilbolton Advanced Meteorological Radar (CAMRa). WesCon – Observing the Evolving Structures of Turbulence (WOEST), will complement WesCon by:  <ul> <li data-leveltext="" data-font="Symbol" data-listid="1" data-list-defn-props="{"335552541":1,"335559685":720,"335559991":360,"469769226":"Symbol","469769242":[8226],"469777803":"left","469777804":"","469777815":"hybridMultilevel"}" aria-setsize="-1" data-aria-posinset="1" data-aria-level="1">capturing the vertical and horizontal spatial variability and evolution of the boundary layer;  </li> <li data-leveltext="" data-font="Symbol" data-listid="1" data-list-defn-props="{"335552541":1,"335559685":720,"335559991":360,"469769226":"Symbol","469769242":[8226],"469777803":"left","469777804":"","469777815":"hybridMultilevel"}" aria-setsize="-1" data-aria-posinset="2" data-aria-level="1">obtaining 3D estimates of small-scale turbulence and larger-scale turbulent coherent structures in convective clouds; and  </li> <li data-leveltext="" data-font="Symbol" data-listid="1" data-list-defn-props="{"335552541":1,"335559685":720,"335559991":360,"469769226":"Symbol","469769242":[8226],"469777803":"left","469777804":"","469777815":"hybridMultilevel"}" aria-setsize="-1" data-aria-posinset="3" data-aria-level="1">supporting the WesCon model evaluation and development, including by providing novel retrievals of turbulent and dynamic processes. </li> </ul> The research will enhance the WesCon campaign by observing turbulent and dynamic processes at sub-km length scales and at a frequency down to 2 minutes or less, enabling a process-oriented evaluation of moist convection and boundary-layer evolution. The approach combines a pair of steerable research-grade Doppler radars with innovative adaptive scanning strategies that enable scanning the same convective cloud targeted by the FAAM aircraft at 2-minute intervals. Two additional X-band Doppler radars will provide excellent coverage of the research domain in southern England, capturing the 3D evolution of convective-cloud dynamics at 5-minute intervals. An additional supersite at Chilbolton will host additional instruments including a new Raman lidar, UHF radar wind profiler and a new array of cloud cameras, to study turbulence in the boundary layer and its influences on convective cloud development.   Through supporting two mobile radiosonde stations, and drone capability, analysis of moist convective turbulence and boundary-layer evolution will be placed in the context of variability in the pre-convective environment. The observations will lead to new understanding of the variability of turbulence and cloud dynamics from 10m-1km scales and of the variability and evolution of the boundary layer in the context of the surrounding cloud field, both of which will enable new evaluation approaches for sub-km grid-length numerical weather prediction (NWP) models.  Department of Computer Science, University of Oxford Projects All oucl-All-WOEST Wed, 13 May 2026 19:06:59 GMT Real World Benchmarks for Deep Probabilistic AI http://www.cs.ox.ac.uk/projects/RealWorldBenchmarksforDeepProbabilisticAI/ Bayesian deep learning (BDL) is an emerging sub-field of Deep Probabilistic AI which stands at the core of probabilistic programming languages such as Edward. BDL offers a pragmatic approach to combining Bayesian probability theory together with deep learning models in practical and scalable ways, giving tools to quantify what deep models “know”. Even though the number of applications making use of BDL might be increasing quickly, the development of the field itself is impeded by the lack of realistic benchmarks to guide research. Evaluating new inference techniques on real applications requires expert domain knowledge, and instead currently researchers developing new inference tools for BDL often use MNIST-like toy benchmarks, ignoring cost of development or scalability aspects. To make significant progress in the deployment of BDL and new deep AI inference tools, new tools must scale to real world settings, and for that researchers must be able to evaluate their inference and iterate quickly with real world benchmark tasks.  This project aims to develop a set of real-world probabilistic-programming-language-agnostic evaluations and data to benchmark probabilistic AI inference techniques, and BDL in particular. It will do this by refining applications which already make use of BDL, and develop additional benchmarks together with Intel Labs, making use of Intel Xeon servers. Additionally, the project will include the designing of a public competition with the developed benchmarks to be hosted at the Bayesian Deep Learning workshop at NIPS (Neural Information Processing Systems) Conference.   The benchmarks will make testing new inference techniques radically easier, leading to rapid development of new tools. With the community competing on the benchmarks, this should lead to significant advancements in reliability of existing and new deep probabilistic AI tools in real world applications.  Department of Computer Science, University of Oxford Projects All oucl-All-Real World Benchmarks for Deep Probabilistic AI Wed, 13 May 2026 19:06:59 GMT Modernising Electrochemical Enzymology to Map Electron Transfer (Enzyme e-map) http://www.cs.ox.ac.uk/projects/Enzymee-map/ In Nature the production of hydrogen and methane fuel molecules from readily available starting materials such as water and carbon dioxide is achieved selectively, efficiently and rapidly by electrocatalytic redox-metalloenzymes containing non-precious transition metal active sites.   The outstanding recent scientific advances made in molecular biology have made the development of biofuel technologies based on these enzymes a reality, but such applications require a complementary toolkit of physical chemistry methods that can dissect how DNA sequence and protein structure relates to function.   Classic bio-electrochemistry methods developed in the 1980s have been a powerful way to probe the active site reactivity of such enzymes, but they have been unable to map the electron transfer processes which underpin the catalysis. Therefore, we have been limited to a narrowly active-site focussed view of enzyme mechanism.   This project will transform the state of the art in bio-electrochemistry to deliver a powerful new technique that can “see” the electron-transfer processes of the highly evolved and essential electron-transfer reaction centres in redox-enzymes, and deconvolute their role in electrocatalysis.   This will be achieved by deploying advanced computational methods to integrate intelligent experimental design into electrochemistry to develop a methodology that enables the separation and accurate modelling of the electron transfer processes of an enzyme bound to substrate, and chemical biology methods to develop linker molecules for light-activated electrografting of proteins and enzymes onto electrodes.   The project will showcase the power of this new electrochemical enzymology toolkit by conducting previously impossible hypothesis-led investigations and enzyme-discovery projects into i) cellulose-degrading LPMOs that play a crucial role in biorefinery enzyme cocktails and ii) hydrogenases, Ni+Fe or Fe-only metalloenzymes that are as rapid and efficient at hydrogen-catalysis as platinum.  The project is a collaboration with the University of York (lead).   Department of Computer Science, University of Oxford Projects All oucl-All-Enzyme e-map Wed, 13 May 2026 19:06:59 GMT The Large Agent Collider http://www.cs.ox.ac.uk/projects/TheLargeAgentCollider/ The aim of this project is to effect a step change in the ability to develop and deploy valid and robust, large-scale, agent-based models.   Agent-based models are increasingly used throughout industry and academia, in areas ranging from financial modelling to logistics and supply chain management, where they are used to model complex socio-technical systems down at the level of individual actors.  Agent-based models allow the capturing of aspects of systems (such as emergent properties, which arise in unpredictable ways from the interaction of many agents) that conventional modelling does not permit. Agent-based modelling came to international prominence when an agent-based epidemiological model of COVID-19 was revealed as one of the key drivers behind the UK government's decision to enter a lockdown in March 2020.   Although they are widely used, as an engineering discipline agent-based modelling remains in its infancy, and subsequent criticisms of the COVID-19 model highlight common difficulties in agent-based models. First, current agent-based modelling environments force the embedding of key assumptions directly within code, thereby obfuscating such assumptions and making it hard to understand them (clearly essential for situations such as the COVID model). Second, there needs to be better ways of populating such models with realistic agent behaviours. Third, such models are limited in the extent to which their predictions can be relied upon: knowledge is lacking as to how to calibrate such models. Fourth, there is no available methodology for validating such models: existing techniques (e.g., model checking, used for formally verifying that systems satisfy their requirements) are unsuitable in their present form for agent-based models.  Using state-of-the-art techniques in AI and machine learning, this project will see fundamental research in the development of the scientific and engineering methodology necessary to transform capability with respect to modelling, populating, calibrating, and validating agent-based models at scale. Working with industrial partners, and building on extensive previous in-house models, techniques will be tested and refined on a range of case studies. If successful, agent-based modelling will be transformed from an ad hoc, trial and error process into a robust engineering discipline with a rigorous methodological foundation. The project will establish Oxford as a world leader in the applications and analysis of multi-agent systems in general, and agent-based modelling specifically, and will greatly strengthen the UK’s capabilities in this important and rapidly expanding area.  Department of Computer Science, University of Oxford Projects All oucl-All-The Large Agent Collider Wed, 13 May 2026 19:06:59 GMT Beyond Linear Dynamical Systems http://www.cs.ox.ac.uk/projects/BeyondLinearDynamicalSystems/ Dynamical systems pervade the quantitative sciences, e.g., recurrence sequences appear across computer science, combinatorics, number theory, economics, and theoretical biology, among many other areas. Such systems are typically simple to describe, yet they have a rich algorithmic and mathematical theory that is replete with simply stated and compelling open problems.   The goal of this project is to achieve a major advance in the algorithmic theory of fundamental dynamical systems arising in verification and related areas. The research programme aims to expand the frontier of what can be decided on recurrence sequences, piecewise affine maps, constraint loops, linear time-invariant systems, and numeration systems. The decision problems that will be considered (reachability, invariant synthesis, model checking, etc.) are particularly relevant to algorithmic verification, automata theory, and program analysis.   The project represents an ambitious advance, with a major rationale being to prosecute new approaches to central open problems on linear systems, such as the Skolem Problem. In addition, it aims to go beyond pure linear dynamics and analyse models with conditional branching, nondeterminism, an external controller, and polynomial recursivity. To this end, the methodology combines automata theory, model theory, and symbolic dynamics, on the one hand, with algebraic and Diophantine geometry, on the other.   If successful, this project will make significant progress on longstanding open problems and will open new lines of research at the boundary of computer science and mathematics.  Department of Computer Science, University of Oxford Projects All oucl-All-Beyond Linear Dynamical Systems Wed, 13 May 2026 19:06:59 GMT Proof complexity and circuit complexity: a unified approach http://www.cs.ox.ac.uk/projects/Proofcomplexityandcircuitcomplexity:aunifiedapproach/ Can we automate creative and challenging tasks such as proving mathematical theorems or designing learning algorithms? Such questions can be formalised in the language of computational complexity theory and constitute some of the most fundamental scientific problems of our times. A famous obstacle in the centre of these pursuits, known as the P versus NP problem, asks whether it is possible to solve efficiently all problems whose solutions can be efficiently verified. A positive answer to the question would single-handedly resolve a myriad of major problems including theorem-proving and algorithm-design.  Unfortunately, the P versus NP problem has resisted solution for over 50 years. In fact, it is consistent with our current knowledge that almost all problems in practice are solvable extremely efficiently with very small computational devices. The decades of research have led to two prominent but contrasting perspectives on the problem: the meta-mathematical view manifested in the theory of complexity of proofs (proof complexity) and a concrete combinatorial view dominant in the analysis of computational models such as logical circuits (circuit complexity).  This project focuses on bridging these approaches. It will build on the emerging theory of hardness magnification and a rise of related new connections between logic, learning theory and cryptography. It thus has the potential to bring us closer to a full understanding of the power of computation and logical reasoning.  Department of Computer Science, University of Oxford Projects All oucl-All-Proof complexity and circuit complexity: a unified approach Wed, 13 May 2026 19:06:59 GMT FAIR: Framework for Responsible Innovation for Responsible Adoption of AI in the Financial Services Industry http://www.cs.ox.ac.uk/projects/FAIR/ The project will achieve a step change in terms of the ability of financial service providers to enable trustworthy data-driven decision-making, while enhancing their resilience, accountability and operational robustness using AI.   AI technologies have the potential to unlock significant growth for the UK financial services sector through novel personalised products and services, improved cost-efficiency, increased consumer confidence, and more effective management of financial, systemic, and security risks. However, there are currently significant barriers to adoption of these technologies, which stem from a capability deficit in translating high-level principles (of which there is an abundance) concerning trustworthy design, development and deployment of AI technologies ("trustworthy AI"). These include safety, fairness, privacy-awareness, security, transparency, accountability, robustness and resilience, to concrete engineering, governance, and commercial practice.  In developing an actionable framework for trustworthy AI, the major research challenge that needs to be overcome lies in resolving the tensions and trade-offs which inevitably arise between all these aspects when considering specific application settings. For example, reducing systemic risk may require data sharing that creates security risks; testing algorithms for fairness may require gathering more sensitive personal data; increasing the accuracy of predictive models may pose threats to fair treatment of customers; improved transparency may open systems up to being "gamed" by adversarial actors, creating vulnerabilities to system-wide risks.  This comes with a business challenge to match. Financial service providers that are adopting AI approaches will experience a profound transformation in key areas of business as customer engagement, risk, decisioning, compliance and other functions transition to largely data-driven and algorithmically mediated processes that involve less and less human oversight. Yet, adapting current innovation, governance, partnership and stakeholder relation management practice in response to these changes can only be successfully achieved once assurances can be confidently given regarding the trustworthiness of target AI applications.  The research hypothesis of this project is based on recognising the close interplay between these research and business challenges:  <ul> <li data-leveltext="" data-font="Symbol" data-listid="1" data-list-defn-props="{"335552541":1,"335559685":720,"335559991":360,"469769226":"Symbol","469769242":[8226],"469777803":"left","469777804":"","469777815":"hybridMultilevel"}" aria-setsize="-1" data-aria-posinset="1" data-aria-level="1">Notions of trustworthiness in AI can only be operationalised sufficiently to provide necessary assurances in a concrete business setting that generates specific requirements to drive fundamental research into practical solutions, with solutions which balance all of these potentially conflicting requirements simultaneously. </li> <li data-leveltext="" data-font="Symbol" data-listid="1" data-list-defn-props="{"335552541":1,"335559685":720,"335559991":360,"469769226":"Symbol","469769242":[8226],"469777803":"left","469777804":"","469777815":"hybridMultilevel"}" aria-setsize="-1" data-aria-posinset="2" data-aria-level="1">Recognising the importance of close industry-academia collaboration to enable responsible innovation in this area, the partnership will embark on a systematic programme of industrially driven interdisciplinary research, building on the strength of the existing Turing-HSBC partnership. </li> </ul> The project objectives will be achieved by improving understanding of sequential data-driven decision making, privacy- and security- enhancing technologies, methods to balance ethical, commercial, and regulatory requirements, the connection between micro- and macro-level risk, validation and certification methods for AI models, and synthetic data generation.  The project will help drive innovation across the industry in a safe way, which will help establish the appropriate regulatory and governance framework, and a common "sandbox" environment to enable experimentation with emerging solutions and the testing of their viability in a real-world business context. It will also provide the cornerstone for impact anticipation and continual stakeholder engagement in the spirit of responsible research and innovation.  See the main FAIR project page: <a rel="noopener" href="https://www.turing.ac.uk/research/research-projects/fair-framework-responsible-adoption-artificial-intelligence-financial" target="_blank">https://www.turing.ac.uk/research/research-projects/fair-framework-responsible-adoption-artificial-intelligence-financial</a> Department of Computer Science, University of Oxford Projects All oucl-All-FAIR Wed, 13 May 2026 19:06:59 GMT Optimisation for Game Theory and Machine Learning http://www.cs.ox.ac.uk/projects/OptimisationforGameTheoryandMachineLearning/ Algorithms for optimisation often in practice use local search approaches. For example, when the objective function is continuous and smooth, gradient descent is usually used (for example, in neural networks). In game-theoretic settings, local search arises naturally in the context of multiple agents, who are attempting to improve their payoffs by best-responding to their peers’ behaviour. Of course, there is no general guarantee about the convergence of this process, it depends on the structure of the game. Local search often works surprising well in practice (even when worst-case is known to be poor) and it is of interest to understand why. This project aims to develop the theory of what is going on, and hopefully lead to improved algorithms. It will consider various specific optimisation problems in more detail, as part of this agenda.  Both machine learning and game theory have given rise to diverse problems of local optimisation, and it is of interest to classify them according to their computational hardness. The project aims to study a general issue, which is the “hard in theory, easy in practice” phenomenon of these problems (so, an aspect of the “beyond worst-case analysis” agenda). The project will include the designing of novel algorithms with performance guarantees. In settings of continuous optimisation, where gradient descent is applicable, there are new and interesting variants of the technique, for example ‘optimistic’ gradient descent, and the issue of how to adjust the step size, or learning rate, as the algorithm runs.  Department of Computer Science, University of Oxford Projects All oucl-All-Optimisation for Game Theory and Machine Learning Wed, 13 May 2026 19:06:59 GMT Driving Behaviour in Multi-Winner Elections http://www.cs.ox.ac.uk/projects/BMW/ The long-term goal of this project is to enable groups of strategic users (either human or artificial) to reach high quality, mutually agreed decisions when selecting multiple alternatives.  Modern societies often need to make choices based on the desires and preferences of multiple stakeholders. Such choices range from traffic policies in a local neighbourhood to joining or leaving major political or economic alliances. Similar challenges are faced by many organisations, both commercial and non-profit. Examples include hiring decisions, identifying strategic priorities, and budget allocation. Likewise, independent artificial agents interacting in a common environment may need to agree on a joint plan of action or allocation of resources.  Historically, such scenarios were analysed using the methodology of social choice - a discipline that combines tools of mathematics, economics and political science. More recently, it became clear that one also needs to consider algorithmic aspects of the proposed solutions, which lead to the emergence of the field of computational social choice (COMSOC).  While much of the early COMSOC research considered the setting where the goal is to elect a single winning alternative based on voters' preferences over the alternatives, more recently the focus has shifted to the multi-winner voting setting, where one aims to select k alternatives (a committee). The applications of this model include electing political leaders, shortlisting applicants for jobs or talent competitions, creating portfolios or identifying items to recommend to a user of online media based on other users' experiences, etc.   An even more general setting is that of Participatory Budgeting (PB) - the task of aggregating the voters' preferences to select a subset of projects to implement from a list of options, where each project has a cost and the total cost should not exceed a given budget. PB was initiated in Brazil in 1989 and was envisioned as a way for local residents to allocate public funds in their neighbourhood. Over the next few decades, it quickly spread across the world. For example, in 2022, the city of Paris will allocate over 75 million Euros for urban development by means of PB. PB can capture a variety of applications other than urban planning, such as deciding on a set of measures to achieve a particular target (such as reducing carbon emissions or controlling viral transmission) or allocating the programmers' time in an open-source software community.  Both multi-winner voting and PB have received a lot of attention from the COMSOC community, with researchers identifying general principles for selecting good solutions (axioms) and proposing (computationally efficient) voting rules that satisfy these axioms (or proving impossibility/hardness results). However, much of the existing work assumes that the voters have a complete knowledge of their preferences and report them truthfully. Neither assumption is fully realistic. Voters may have a hard time making up their minds concerning complex proposals (such as evaluating risk and benefits of different energy sources or implementing educational reforms), and they can misreport their preferences if they can benefit from doing so.   The primary focus of this project is to develop a systematic understanding of strategic behaviour in multi-winner voting and PB, with a focus on the associated algorithmic challenges. Specifically, it will evaluate the quality of stable outcomes of strategic voting and establish the complexity of computing them, as well as analyse the dynamics of iterative voting. It will also examine the incentives associated with agents delegating their decisions to more knowledgeable agents. Broadly, the project aims to identify tools for collective decision-making that can drive voting behaviour to desirable outcomes and perform well in realistic settings - i.e., in the presence of uncertainty and bounded rationality. Working with project partners, these results will be applied in practical decision-making scenarios in the contexts of urban living and distributed autonomous organisations.  Department of Computer Science, University of Oxford Projects All oucl-All-BMW Wed, 13 May 2026 19:06:59 GMT The Automatic Computer Scientist http://www.cs.ox.ac.uk/projects/TheAutomaticComputerScientist/ This project aims to significantly advance inductive logic programming so that it can discover novel and complex algorithms and implement large programs without the need for strong human guidance.  Algorithms are ubiquitous: they track our sleep, find us cheap flights, and even help us see black holes. However, designing novel algorithms is extremely difficult, and we do not have efficient algorithms for many fundamental problems. This project aims to accelerate algorithm discovery by building an automatic computer scientist (AutoCS).   The idea of developing machines that automatically write computer programs is a long-standing grand challenge in Artificial Intelligence (AI) and offers the potential to automatically build bug-free and efficient programs. To work towards this, this project will build on major recent breakthroughs in Inductive Logic Programming (ILP), a form of symbolic machine learning based on mathematical logic.  Due to major recent breakthroughs, ILP currently has the ability of a first-year computer science student: given much guidance, it can learn simple algorithms and implement small programs. This project aims to significantly advance ILP to the level of a computer science PhD student so that given little guidance it can discover novel and complex algorithms and implement large programs.  As a marker of its success, a key objective of this project is to use an AutoCS to discover a novel algorithm and publish it in a computer science journal. Such a result would be a landmark achievement for AI and would herald a new era of automatic scientific discovery.  Department of Computer Science, University of Oxford Projects All oucl-All-The Automatic Computer Scientist Wed, 13 May 2026 19:06:59 GMT SMASH-HCM http://www.cs.ox.ac.uk/projects/SMASH-HCM/ Hypertrophic cardiomyopathy (HCM) is the most common inherited heart disease (prevalence 1:200 - 1:500), manifested by thickening of cardiac walls, increasing risks of arrhythmia, and sudden cardiac death. HCM affects all ages - it is the leading cause of death among young athletes. Comorbidities due to gene mutations include altered vascular control, and, caused by HCM, ischemia, stroke, dementia, or psychological and social difficulties. Multiple causal mutations and variations in cellular processes lead to highly diverse phenotypes and disease progression. However, HCM is still diagnosed as one single disease, leading to suboptimal care.   SMASH-HCM will develop a digital-twin platform to dramatically improve HCM stratification and disease management, both for clinicians and patients. Multilevel and multiorgan dynamic biophysical and data-driven models are integrated in a three-level deep phenotyping approach designed for fast uptake into the clinical workflow.   The project will unite 8 research partners, 3 hospitals, 3 SMEs, and a global health-technology corporation, in collaboration with patients, to advance the state of the art in human digital twins: including in-vitro tools, in-silico from molecular to systemic level models, structured and unstructured data analysis and explainable artificial intelligence - all integrated into a decision support solution for both healthcare professionals and patients.   SMASH-HCM will deliver new insights into HCM, improved patient care and guidance, validated preclinical tools, and above all, a first HCM stratification and management strategy, validated in a pilot clinical trial and tested with end users, thus providing a cost efficient and effective solution for this complex disease. It will develop a strategy towards fast regulatory approval and, in reaching its goals, will serve as a basis for future digital-twin platforms for other cardiac diseases integrating models and data from various scales and sources.  Department of Computer Science, University of Oxford Projects All oucl-All-SMASH-HCM Wed, 13 May 2026 19:06:59 GMT Erlangen AI Hub: Mathematical Foundations of Intelligence http://www.cs.ox.ac.uk/projects/AIHub/ View our website at <a rel="noopener" href="https://erlangenhub.ox.ac.uk/" target="_blank">https://erlangenhub.ox.ac.uk/</a> The Erlangen AI Hub exists to revolutionise the application of modern mathematics to understand AI, unifying and expanding the field to unlock new, more intelligent systems. The EPSRC-funded research programme, partnered by industry, brings together leading minds from across the UK’s mathematical, algorithmic and computational communities. It employs foundational tools to break new ground in AI, and redefine its future use to benefit science, industry, the economy and society.  The Erlangen model In 1872, Felix Klein published his Erlangen Programme, which brought a revolutionary and unifying perspective to geometry via symmetries formalised by algebra. With a growing understanding of the geometric nature of data and intelligence, the Erlangen Programme for AI draws inspiration from Klein’s work. It will galvanise new AI technologies based on solid foundations of pure mathematics, harnessing the power of classical theories and encouraging new ones, in mathematics and beyond, to empower the next generation of AI. Building a community Led by Oxford, the hub network consists of six academic institutions spanning the UK, bringing together leading researchers in mathematics, algorithms, and computing. The research programme aims to remove barriers between fields and unify a diverse cohort, exploiting tools from currently underexplored mathematical fields to understand and advance AI. The hub also aims to attract theoreticians to new problems and applications in AI in both scientific and industrial settings.  Advancing AI research The world-leading research at the Erlangen AI Hub will apply powerful ideas from geometry, topology, and other branches of modern mathematics to provide rigorous solutions to four key areas that underlie modern AI systems:   <ul> <li data-leveltext="" data-font="Symbol" data-listid="1" data-list-defn-props="{"335552541":1,"335559685":720,"335559991":360,"469769226":"Symbol","469769242":[8226],"469777803":"left","469777804":"","469777815":"hybridMultilevel"}" aria-setsize="-1" data-aria-posinset="1" data-aria-level="1">Understanding Data: Introduce mathematical methodologies for discovering and expressing hidden structures in data, which then can be exploited by new machine learning models.    </li> </ul> <ul> <li data-leveltext="" data-font="Symbol" data-listid="2" data-list-defn-props="{"335552541":1,"335559685":720,"335559991":360,"469769226":"Symbol","469769242":[8226],"469777803":"left","469777804":"","469777815":"hybridMultilevel"}" aria-setsize="-1" data-aria-posinset="1" data-aria-level="1">Understanding Machine Learning Models: Characterise machine learning models mathematically to understand their success and failure, and ensure their robustness, reliability, and fairness.    </li> </ul> <ul> <li data-leveltext="" data-font="Symbol" data-listid="3" data-list-defn-props="{"335552541":1,"335559685":720,"335559991":360,"469769226":"Symbol","469769242":[8226],"469777803":"left","469777804":"","469777815":"hybridMultilevel"}" aria-setsize="-1" data-aria-posinset="1" data-aria-level="1">Understanding Learning: Use mathematics to understand learning and optimisation algorithms and enable them to benefit from structures in machine learning models and data. </li> </ul> <ul> <li data-leveltext="" data-font="Symbol" data-listid="3" data-list-defn-props="{"335552541":1,"335559685":720,"335559991":360,"469769226":"Symbol","469769242":[8226],"469777803":"left","469777804":"","469777815":"hybridMultilevel"}" aria-setsize="-1" data-aria-posinset="2" data-aria-level="1">Understanding Decision-Making: Enable the building of self-adaptive, largely autonomous AI systems that understand their limitations, increasing reliability and minimising human intervention.  </li> </ul> From science to society The Erlangen AI Hub is one of nine AI research hubs across the UK funded by EPSRC. The hub sits at the centre of a collaborative network of stakeholders, fusing academic and industry knowledge with real world action, bringing world-leading research to applied settings. It aims to achieve rapid and enduring impact in science, industry, government and beyond.    Academic leadership team Hub Directors Professor Michael Bronstein at the University of Oxford, Professor Anthea Monod at Imperial, and Professor Jeffrey Giansiracusa at Durham University will be supported by node leads Professor Jacek Brodzki at University of Southampton, Primoz Skraba at Queen Mary University London, and Professor Ran Levi at University of Aberdeen.  Department of Computer Science, University of Oxford Projects All oucl-All-AI Hub Wed, 13 May 2026 19:06:59 GMT Geometric and Topological ML for Next-generation Drugs and Food http://www.cs.ox.ac.uk/projects/GeometricandTopologicalMLforNext-generationDrugsandFood/ The aim of this project is to develop a novel mathematical framework for geometric and graph machine learning, and apply these new methods to some of the most challenging problems in the domains of drug and food design. This approach will overcome the limitations of existing machine learning (ML) methods and enable a quantitative and qualitative leap that will lead to new capabilities.  Over the past decade, AI and ML methods have had a revolutionary impact in several fields, adding billions in business value, creating new markets, and transforming entire industrial segments. At the same time, in other fields such as medicine and drug design, the hopes for a similar fast impact of AI have not yet materialised. One of the key challenges for AI in the next decade is delivering on these unfulfilled promises and addressing notoriously hard problems in the biomedical and physical sciences, where an AI-driven breakthrough can have a dramatic economic and societal impact. Achieving this goal requires developing a new generation of AI methods that meaningfully exploit domain-specific knowledge, have performance guarantees, are interpretable, and address the needs and concerns of domain experts and the broader society.  The past few years have seen the emergence of "Geometric ML" approaches leveraging broad mathematical principles of symmetry and invariance. Geometric ML allows deriving from first principles the majority of modern ML architectures and also provides a general blueprint to incorporate domain-specific inductive biases in a mathematically principled way, paving the way for future ML systems. Instances of Geometric ML such as Graph Neural Networks (GNNs) and equivariant neural networks have brought a series of breakthroughs in applications ranging from particle physics and fake news detection to pure mathematical proofs and molecule design. The triumph of Geometric ML in biological sciences is perhaps best exemplified by the ground-breaking DeepMind AlphaFold 2 model for protein structure prediction based on geometric equivariant attention. In these and other problems, Geometric ML allows to reason at the right level of abstraction, leading to computationally tractable and at the same time physically correct models.  The vast majority of today's GNNs rely on the message passing paradigm, where graph representation is formed by an exchange of information between graph nodes connected by edges. This "node-and-edge"-centric mindset constitutes a major limitation of current Geometric and Graph ML schemes. From a theoretical viewpoint, message passing is equivalent to iterative graph isomorphism testing (Weisfeiler-Lehman algorithm). As a result, message passing GNNs have limited expressive power and poorly understood generalisation properties and are disadvantageous in chemical and biological applications. Furthermore, the use of the input graph as a computational device for message passing often leads to bottleneck and over-squashing phenomena and is poorly compatible with existing hardware.  This project aims to use tools from differential geometry, algebraic topology, and differential equations to derive a new methodology for deep learning on graphs. The goal is to replace the computational fabric of GNNs with richer and more suitable structures that will allow us to overcome the current limitations of GNNs. These techniques, so far insufficiently explored in the field of ML, will lead to a new generation of geometric and graph machine learning models that are better interpretable, have guarantees of expressive power and performance, are more efficient in the required amount of data and compute, and better exploit existing hardware.  Deep learning models based on these mathematical foundations will be developed into efficient and scalable software implementation and, in collaboration with industrial partners, applied to some of today's most important and challenging problems from the domains of drug and food design. In the long perspective, the new methods are expected to contribute to accelerated drug development, mapping the "dark matter" of food-based bioactive molecules to help cure cancer, and create non-meat alternative foods to reduce the impact of traditional food industries on the global climate.  Department of Computer Science, University of Oxford Projects All oucl-All-Geometric and Topological ML for Next-generation Drugs and Food Wed, 13 May 2026 19:06:59 GMT New Approaches to Approximability of Satisfiable Problems http://www.cs.ox.ac.uk/projects/NAASP/ This project was awarded as an ERC Consolidator Grant in March 2022 as the only ERC CoG awarded to the University of Oxford across all disciplines and the only ERC CoG awarded in Computer Science and Informatics (PE6 panel) in the UK in the 2021 round. Due to the non-association of the UK to the ERC, this grant was transformed into an equivalent UKRI ERC Guarantee Grant. The project runs from 1 July 2022 through 30 June 2027 in the Department of Computer Sciecne at the University of Oxford. Department of Computer Science, University of Oxford Projects All oucl-All-NAASP Wed, 13 May 2026 19:06:59 GMT Better Languages for Statistics http://www.cs.ox.ac.uk/projects/BLAST/ BLAST is an ERC funded project (consolidator grant) that will run from Oct 2020 until Sept 2026. Probabilistic programming is a powerful method for Bayesian statistical modelling, particularly where the sample space is complex or unbounded (non-parametric). This is because the statistical model can be described clearly in a way that is precise but separate from inference algorithms. It accommodates complex models in such a way that outcomes are still explainable. The objective of the proposed research is to develop a semantic foundation for probabilistic programming that properly explains the non-parametric aspects, particularly the symmetries that arise there. There are three ultimate goals: <ul> <li>to propose new probabilistic programming languages: better languages for statistics;</li> <li>to devise new general inference methods for probabilistic programs;</li> <li>to build new foundations for probability.</li> </ul> The method is to build on advances on exploiting symmetries in traditional programming language semantics, by combining this with recent successes in formal semantics and verification for probabilistic programming. Department of Computer Science, University of Oxford Projects All oucl-All-BLAST Wed, 13 May 2026 19:06:59 GMT Securely Fusing Cooperative and Non-Cooperative Data for Maritime Domain Awareness http://www.cs.ox.ac.uk/projects/SecurelyFusingCooperativeandNon-CooperativeData/ This research project aims to enhance maritime domain awareness (MDA) by securely integrating automatic identification system (AIS) data with synthetic aperture radar (SAR) satellite imagery. The project seeks to improve the detection and classification of vessels, reducing reliance on manual intervention. By addressing security vulnerabilities in AIS and limitations in SAR data, the research will contribute to more reliable, automated monitoring of maritime activity.  A significant challenge in MDA is ensuring accurate vessel detection and classification. AIS, a cooperative tracking system, provides frequent location updates but is vulnerable to falsification and non-compliance. Conversely, SAR is a non-cooperative system capable of detecting ships regardless of their willingness to be tracked. However, SAR has drawbacks, including low image resolution and infrequent updates. The inability to securely and effectively fuse these two data sources hampers the development of robust MDA systems.  This project aims to address these challenges by designing a system that systematically analyses security threats to SAR/AIS fusion, mitigating potential attacks and misclassifications. The system will implement attacker-aware ship classification, assigning confidence scores to detections and autonomously determining when additional data is required. Additionally, novel techniques such as radio frequency (RF) transmitter fingerprinting and location verification methods will be developed to authenticate AIS signals and verify their claimed positions, thereby strengthening the integrity of vessel tracking data.  The research will have far-reaching implications for maritime security, particularly in combating illegal activities such as smuggling, piracy, and unregulated fishing. By enhancing confidence in MDA systems, the project will provide governments, law enforcement agencies, and maritime organisations with more reliable tools for monitoring maritime activity. This work builds on the research team’s expertise in security analysis and RF verification, ensuring that the solutions developed are both technically robust and practically implementable.  Department of Computer Science, University of Oxford Projects All oucl-All-Securely Fusing Cooperative and Non-Cooperative Data Wed, 13 May 2026 19:06:59 GMT