University of Birmingham Statistics and Data Science Day

Schedule

Tue Dec 01 2026 at 01:00 pm to 04:30 pm

UTC+00:00
Location

Lecture Theatre G33, Ground Floor, Aston Webb Building | Birmingham, EN

Advertisement
An afternoon of talks spanning statistical theory, computational methods, data science, and real-world applications.
About this Event

Join us at the University of Birmingham for Statistics and Data Science Day, an afternoon bringing together researchers from across the UK to share recent developments in statistical theory, computational methods, data science, and real-world applications.

The programme features four invited speakers from the Universities of Leeds, Lancaster, Bristol, and Liverpool, with talks spanning functional data analysis and brain imaging, computation-aware Gaussian processes, Markov chain Monte Carlo methods, and data-informed epidemic modelling.

The event is aimed at academics, postdoctoral researchers, PhD students and others interested in statistics, data science, probability, machine learning, applied mathematics, and related areas. The talks will showcase both methodological advances and applications, while remaining accessible to participants from a range of backgrounds.

There will also be opportunities throughout the afternoon to meet other researchers, exchange ideas, and make new connections.



Tentative Programme

13:00–13:45 — Haiyan Liu (University of Leeds)
Functional Data Analysis Beyond Curves: Methods and Applications to Brain Imaging

13:45–14:30 — Disha Hegde (Lancaster University)
Computation-aware Gaussian Process Inference: Calibration and Implementation

14:30–15:00 — Coffee break

15:00–15:45 — Sam Power (University of Bristol)
Sharp Contraction Bounds for Kinetic MCMC Samplers in the Log-Concave Setting

15:45–16:30 — Emily Nixon (University of Liverpool)
Data-informed Modelling for Sheep Scab Control



Talks
Haiyan Liu — University of Leeds

Functional Data Analysis Beyond Curves: Methods and Applications to Brain Imaging

Functional Data Analysis (FDA) provides a flexible statistical framework for analysing data that are naturally represented as functions, curves, surfaces, or images. In this talk, I will introduce the fundamental ideas of FDA and review two of its core methodologies: functional principal component analysis (FPCA) and functional linear regression. After outlining the mathematical foundations of these approaches, I will present our recent work on extending FPCA to accommodate more general functional data in the form of images.

These methodological advances are motivated by modern neuroimaging applications, where high-dimensional image data are routinely collected. I will illustrate the proposed methods using functional and structural MRI data from studies of decision-making and Alzheimer's disease, demonstrating how FDA can uncover meaningful patterns and provide interpretable representations of complex imaging data.


Disha Hegde — Lancaster University

Computation-aware Gaussian Process Inference: Calibration and Implementation

Gaussian processes are a powerful and flexible tool for Bayesian nonparametric regression, allowing a user to fit a wide array of possibly nonlinear phenomena using simple computational routines. However, they are notorious for scaling cubically with the size of the training set, preventing application to very large regression problems. Computation-aware Gaussian processes (CAGPs) tackle this scaling issue by exploiting probabilistic linear solvers to approximate an expensive linear solve, widening the posterior with additional computational uncertainty due to reduced computation. This talk introduces CAGPs and discusses the calibration of the uncertainty from the widened posterior. We then discuss choice of underlying probabilistic linear solver, and the resulting convergence-calibration tradeoff. We finally introduce affine tracing, a framework that automatically constructs probabilistic solvers from standard implementations of affine iterative methods, allowing more sophisticated solvers such as multigrid to be used for CAGPs while preserving calibration.


Sam Power — University of Bristol

Sharp Contraction Bounds for Kinetic MCMC Samplers in the Log-Concave Setting

Sampling from a complicated probability distribution is a central task in Bayesian statistics, where it allows us to estimate quantities of interest and quantify uncertainty. Popular sampling algorithms, such as Hamiltonian Monte Carlo, use ideas from particle dynamics to explore these distributions efficiently. A key question is how rapidly these algorithms forget their initial conditions and converge to equilibrium.

In this talk, I will explain how to study this question using the coupling method: running two trajectories with different starting positions but common underlying randomness. For strongly log-concave distributions with controlled curvature, we obtain sharp bounds on how closely the trajectories approach one another, yielding quantitative convergence guarantees. The central observation is that their contraction can be bounded by comparison with the same dynamics applied to suitable isotropic Gaussian targets. This provides a common approach to Hamiltonian Monte Carlo, underdamped Langevin dynamics, and several of their numerical discretisations.

The talk will begin with a gentle introduction to the sampling methods and the coupling viewpoint, before developing the mathematical comparison argument. No prior background in stochastic simulation will be assumed.


Emily Nixon — University of Liverpool

Data-informed Modelling for Sheep Scab Control

Sheep scab is a contagious parasitic disease of sheep with major animal health and economic consequences. In this talk I will describe how outbreak records, sheep movement data, and farm census information can be combined with a mechanistic model to study transmission and evaluate control strategies in Great Britain. The model is a hybrid spatial epidemic framework, with stochastic infection dynamics, environmental contamination, and discrete movement events linking farms. Although the modelling is primarily mechanistic, a key part of the work is inference from incomplete and heterogeneous data, including simulation-based fitting using SMC-ABC.

I will show how this framework has been used to reconstruct past transmission patterns, identify high-risk areas, and assess the likely impact of interventions such as movement controls, targeted treatment, and vaccination. I will also touch on current extensions, including vaccination and farmer behaviour, which highlight the importance of combining mechanistic models with behavioural and operational data. More broadly, the talk will illustrate how mathematical thinking can connect data, uncertainty, and policy in a real-world applied setting.orld applied setting.

Advertisement

Where is it happening?

Lecture Theatre G33, Ground Floor, Aston Webb Building, Edgbaston, Birmingham, United Kingdom

Event Location & Nearby Stays:

Tickets

GBP 0.00

Know what’s Happening Next — before everyone else does.
Xiaocheng Shang
Host or PublisherXiaocheng Shang

Ask AI if this event suits you