Our faculty research specialisms

Mona Azadkia designs nonparametric statistical tools that avoid strong assumptions, helping uncover relationships and independence within complex data. Her work supports more accurate feature selection, stronger hypothesis testing, and applications such as DNA sequence alignment.

Future efforts will target measuring dependence in network-type data, improving independence tests using modern diagnostic approaches, and modelling ranked or partially ranked data common in surveys and recommendation systems.

Selected publications

· Azadkia, M., Balabdaouni, F., Linear regression with unmatched data: a deconvolution perspective, JMLR, 2024

· Azadkia, M., Chatterjee, S., A simple measure measure of conditional independence, Annals of Statistics, 2021

Dr Marcos Barreto is interested in big data linkage and analytical tools, machine learning applied to health and socioeconomic data, and artificial intelligence in data science education. His recent projects centred around anomaly detection for pandemic anticipation, social inequalities indexing, artificial intelligence for COVID-19 monitoring, and a series of educational projects on the use of generative artificial intelligence in data science programming education.

Selected publications

· Borges D. G. F. et al, An integrated framework for modelling respiratory disease transmission and designing surveillance networks using a sentinel index. Royal Society Open Science, 2025

· Silva J. C. et al, Mapping student-GenAI interactions onto experiential learning: the GENIAL framework, Studies in Higher Education, 2025

· Santo J. S. E. et al. K-means DTW Barycenter Averaging: a clustering analysis of COVID-19 cases and deaths on the Brazilian federal units. Journal of Data Science and Analytics, 2025

Dr Yunxiao Chen’s research focuses on psychometrics and advanced statistical models to understand complex patterns in data and support sequential decision-making. His work is applied to large-scale educational assessments, personalised learning, recommendation systems, ranking problems, and data visualisation.

Looking ahead, he aims to develop psychometric methods specifically for evaluating Generative AI systems.

Selected publications

· Moustaki I. et al, Analysis of multivariate social data: statistical machine learning methods, 3^rd Edition, Taylor & Francis, Oxfordshire, UK, 2026

· Chen, Y. et al, Item response theory – a statistical framework for psychological measurement (with discussions), Statistical Science, 2025

· Chen, Y. Li, X., A note on entrywise consistency for mixed-data matrix completion, JMLR, 2024

Angelos Dassios studies Monte Carlo simulation and Bayesian priors for machine learning, focusing on efficient methods for analysing network and high-dimensional data. His work develops both theoretical and computational approaches to posterior sampling in complex hierarchical and nonparametric Bayesian models.

Recent projects include truncated representations of completely random measures, beta processes, and Pitman–Yor hierarchical models, enabling scalable and exact inference for challenging probabilistic systems. His goal is to create tools that improve Bayesian modelling for networks and structured data, bridging advanced statistical theory with practical machine learning applications.

Selected publications

· Zhang, J., Dassios, A., Posterior sampling from truncated Ferguson-Klass representation of normalised completely random measure mixtures, Bayesian Analysis, 2025

· Zhang, J. et al, Truncated inverse-Lévy measure representation of the beta process, JMLR, 2025

· Zhang, J., Dassios, A., Truncated two-parameter Poisson–Dirichlet approximation for Pitman–Yor process hierarchical models, Scandinavian Journal of Statistics, 2024

-

Alessandro De Palma’s research is concerned with efficiently building neural networks that are provably robust, fair, and explainable. Using techniques from optimisation and formal methods, his work is concerned with both formally proving and enforcing trustworthiness properties for deep learning systems. A large body of his work has focussed on guaranteeing that neural networks are robust to semantics-preserving perturbations of their inputs (adversarial attacks).

His research is particularly relevant for applications where formal guarantees on the deployed ML algorithms are desirable or required, including safety-critical systems (for instance, in medicine or within industrial use-cases), with a particular emphasis on computer vision models.

Selected publications

· De Palma, A., Bunel, R., Dvijotham, K., Kumar, P.K., Stanforth, R., Lomuscio, A., Expressive losses for verified robustness via convex combinations, ICLR 2024

· De Palma, A., Behl, H.S., Bunel, R., Torr, P.H.S., Kumar, P.K., Scaling the convex barrier with sparse dual algorithms, JMLR (2024)

· Kurin, V., De Palma A., Kostrikov, I., Whiteson, S,, Kumar, P.K., In defense of the unitary scalarization for deep multi-task learning, NeurIPS 2022

Dr Kalogeropoulos works at the intersection of machine learning, Bayesian statistics, and sequential modelling to understand complex systems in epidemics and financial markets. His research focuses on creating interpretable models that can predict and explain the spread of infectious diseases, track market interest rates and volatility, and support real-time decision-making.

Current projects explore hierarchical and multi-task models, causal inference, and non-parametric factor analysis, with a goal of advancing both scientific understanding and practical applications in public health and finance.

Selected publications

· Bouranis, A. et al, Bayesian analysis of diffusion-driven multi-type epidemic models with application to COVID-19, JRSS A, 2025

· Dubiel-Teleszynski, T., Kalogeropoulos, K., Karouzakis, N., Sequential learning and economic benefits from dynamic term structure models, Management Science, 2024

· Vamvourellis, K., Kalogeropoulos, K., Moustaki, I., Assessment of generalised Bayesian structural equation models for continuous and binary data, British Journal of Mathematical and Statistical Psychology, 2023

Ieva Kazlauskaite focuses on the computational and statistical aspects of Bayesian inverse problems and the optimal design of experiments to address complex challenges in climate and ice sheet modelling. Her current work is centred on predicting Antarctic ice melt and understanding its implications for rising global sea levels.

By combining computational efficiency with rigorous statistical methods, she develops approaches that help interpret climate simulator data and improve the design of experiments for ice sheet studies. Her goal is to provide actionable insights that not only advance our understanding of climate systems but also support science-driven decision-making on climate change mitigation and adaptation strategies.

Selected publications

• Glyn-Davies, A., A primer on variational inference for physics-informed deep generative modelling, Philosophical Transactions A, 2025

• Williams, C. R., et al, Calculations of extreme sea level scenarios are strongly dependent on ice sheet model resolution, Communications Earth & Environment, 2025

• Vadeboncoeur, A. et al, Random grid neural processes for parametric partial differential equations, ICML, 2023

Clifford Lam studies high-dimensional and spatio-temporal time series to uncover patterns in financial, economic, and epidemiological data. His current work focuses on understanding spatial contagion, such as how shocks spread across markets, economies, or populations, and improving forecasting in complex time series.

By combining tensor analysis, spatial econometrics, and large covariance modelling, I develop methods that reveal hidden structures in interconnected systems. Future work aims to extract common indices from complex datasets, enhance predictions from large-scale data, and detect change-points in spatial spillovers, providing insights that inform both research and real-world decision-making.

Selected publications

· Cen, Z., Lam, C., Tensor time series imputation through tensor factor modelling, Journal of Econometrics, 2025

· Chen, W., Lam, C., Rank and factor loadings estimation in time series tensor factor model by pre-averaging, Annals of Statistics, 2024

· Lam, C., Cen, Z., Matrix-valued factor model with time-varying main effects, Journal of Econometrics, 2024

Giulia Livieri explores the mathematics of machine learning, dynamical systems, and finance, with a focus on mean field games and financial econometrics. Her current work includes approximating conditional laws of stochastic processes, building universal causal deep learning models, analysing bank leverage dynamics, and applying graph-based learning to complex systems.

By combining mathematical rigor with computational methods, Giulia Livieri aims to improve understanding and prediction in financial systems, sustainable finance, and AI applications. Future work will focus on machine learning for mean field games, the foundations of graph learning, and advancing insights into large language models, bridging theory and practical applications.

Selected publications

· Galimberti, L., Kratsios, A., Livieri, G., Designing universal causal deep learning models: The case of infinite-dimensional dynamical systems from stochastic analysis, Constructive Approximation, 2025

· Gambara, M., Livieri, G., Pallavicini, A., Machine-learning regression methods for American-style path-dependent contracts, Quantitative Finance, 2025

· Lillo, F. et al, Analysis of bank leverage via dynamical systems and deep neural networks, SIAM Journal on Financial Mathematics, 2023

Motivated by the ever-increasing prevalence of ML systems and their potential role in high-stakes decisions, the work of Dr Joshua Loftus is concerned by the development of ML algorithms that are fair and interpretable. While fairness ensures that ML systems do not propagate or amplify biases present in data, interpretability is crucial for stakeholders to understand, evaluate, and challenge decisions made by automated systems. His research analyses these properties of AI systems through the lens of causal methods, which capture and account for causal pathways within data.

Joshua Loftus is a vocal proponent of a human-centered approach to AI, and is interested in responsible AI at large. His research enjoys potential applications across disciplines where ML predictions directly impact people, including hiring, credit scoring, healthcare, social policy, and more.

Selected publications

· Loftus, J., Bynum, L., Hansen, S., Causal dependence plots, NeurIPS 2024

· Loftus, J., Position: The causal revolution needs scientific pragmatism, ICML 2024

· Bynum, L., Loftus, J., Stoyanovich, J., Counterfactuals for the future, AAAI 2023

Chengchun Shi’s work lies at the intersection between AI and statistics, with a strong focus on Reinforcement Learning (RL) and, more recently, Large Language Models (LLMs). His work, which was the recipient of an EPSRC grant on “Statistical Methods in Offline Reinforcement Learning” applies rigorous statistical methods to study how agents can learn and interact from their environment to maximise rewards under uncertainty: a paradigm that has become fundamental for the fine-tuning of state-of-the-art LLMs.

Chengchun is interested in LLM trustworthiness and accountability, and has published articles and software applied towards mobile health, neuroscience, video-sharing and ride-sharing.

Selected publications

· Xu, E., Ye, K., Zhou, H*., Zhu, L., Quinzan, F. and Shi, C., Doubly Robust Alignment for Large Language Models, NeurIPS 2025

· Behnamnia, A., Aminian, G., Aghaei, A., Shi, C., Tan, V.Y., Rabiee, H., Log-Sum-Exponential Estimator for Off-Policy Evaluation and Learning, NeurIPS 2025 (spotlight)

· Uehara, M., Kiyohara, H., Bennett, A., Chernozhukov, V., Jiang, N., Kallus, N., Shi, C. and Sun, W., Future-Dependent Value-Based Off-Policy Evaluation in POMDPs, NeurIPS 2023 (spotlight)

The research of Zoltán Szabó focuses on statistical machine learning, with particular emphasis on kernel methods and information-theoretical estimators. His research is foundational in nature and is concerned with improving the capabilities (for instance, in terms of applicability or scalability) of wide-reaching and successful fundamental ML algorithms.

The applications of his research include: safety-critical learning, Bayesian inference, finance, economics, analysis of climate data, criminal data analysis, remote sensing, natural language processing, and gene analysis.

Selected publications

· Kalinke, F., Szabó, Z., Sriperumbudur, B., Nyström kernel Stein discrepancy, AISTATS 2025

· Kalinke, F., Szabó, Z., The minimax rate of HSIC estimation for translation-invariant kernels, NeurIPS 2024

· Bonnier, P., Oberhauser, H., Szabó, Z., Kernelized cumulants: Beyond kernel mean embeddings, NeurIPS 2023

Milan Vojnovic studies machine learning and optimisation, focusing on stochastic and online algorithms, multi-armed bandits, and multi-agent systems. His work emphasizes trustworthy AI, including fairness, selection and ranking methods, and scalable computation for large language models.

Applications include online platforms, information systems, and community integrity platforms, where he develops methods that improve both training and inference-time efficiency. Prof. Milan Vojnovic’s current focus is on ensuring AI systems behave appropriately, combining algorithmic rigor with ethical and practical considerations to create reliable, fair, and scalable AI solutions.

Selected publications

· Tax, N. et al, McGrad: Multicalibration at web-scale, ACM KDD 2026

· Kim, J., Vojnovic, M., Yun, S., An Adaptive Approach for Infinitely Many-armed Bandits under Generalized Rotting Constraints, NeurIPS 2024

· Haimovich, D. et al, On the Convergence of Loss and Uncertainty-based Active Learning Algorithms, NeurIPS 2024

Qiwei Yao studies complex time series, dynamic networks, and spatio-temporal data, combining factor models, dimension reduction, and deep learning to uncover patterns in high-dimensional, interconnected systems. His current work focuses on electricity load forecasting with EDF, developing models that improve prediction and inform operational decisions.

Looking ahead, Qiwei aims to advance deep nonlinear analysis for time series, explore nonlinear cointegration, and forecast future time series using multi-modal data, bridging statistical rigor with modern machine learning to tackle real-world temporal and network challenges.

Selected publications

· Jiang, B. et al, A two-way heterogeneity model for dynamic networks, The Annals of Statistics, 2025

· Han, Y. et al, Simultaneous decorrelation of matrix time series, JASA, 2024

· Jiang, B., Li, J. and Yao, Q., Autoregressive networks, JMLR, 2023

Tengyao Wang studies high-dimensional statistics, change-point analysis, and representation learning to uncover patterns in complex datasets. His current work focuses on developing methods for robust and interpretable learning with applications in bioinformatics, medical research, financial data analysis, and material discovery.

Looking ahead, Tengyao aims to advance representation learning for pairwise comparisons, adaptive conformal model selection, and complex change-point detection via autoencoders, with a particular focus on improving Paediatric Early Warning systems, bridging cutting-edge statistics with real-world impact.

Selected publications

• Ma, T., Wang, T. and Samworth, R. J., Deep learning with missing data, arxiv, 2025

• Yang, X. and Wang, T., Multiple-output composite quantile regression through an optimal transport lens, COLT, 2024

• Jie, L. et al, Automatic change-point detection in time series via deep learning, JRSS B, 2024