Stephen J. Roberts

University of Oxford
Country flagUnited Kingdom

Research Interests

ApplyKite TwinLearns as you apply

Make your next move.

Prepare your application

Review drafts before you send them.

Explore related searches

Articles (5)

This study evaluates the response method for predicting tidal currents. We introduce a coupled response model which explicitly accounts for interactions between velocity components. By leveraging non‐parametric and data‐driven weight estimation, the approach demonstrates superior predictive accuracy compared to classical harmonic analysis (HA), particularly for fast‐moving and non‐linear tidal currents. Using ADCP data from the world's largest deployment of tidal stream turbines, the coupled model achieves superior accuracy with fewer than 30 days of input measurements compared to HA using over 180 days of data. Accuracy improvements extend to both current predictions and the derived harmonic constituents, obtained through a specialized procedure. The response approach shows greater robustness when applied to extremely sparse data. This is reflected by the pseudo‐admittances, which also show the non‐parametric approach advanced can effectively capture unsmooth deviations in the admittance. Analysis of 40 active NOAA current stations highlight when the response approach should and should not be used, yielding average reductions in absolute error of 9.6%. The framework offers new opportunities for studying non‐tidal forcing and sediment transport and has significant implications for tidal energy site development. The proposed method is implemented in the open‐source RTide Python package, providing a practical and accessible tool that reduces the level of expertise required to apply the response method to higher‐order nonlinear processes.

Year:

2025

The volume and complexity of scientific literature are expanding rapidly, making it increasingly difficult to extract and synthesise information across studies. This chal-lenge is particularly evident in the biological sciences, where data span from molecules to ecosystems and linking across hierarchical levels is inherently challenging. Large Language Model (LLM) pipelines offer a scalable solution, but most lack modular-ity, transparency, and mechanisms for human oversight. We present MetaBeeAI, an open-source, modular pipeline that uses LLMs to extract structured information from scientific papers for systematic review and meta-analysis. The system includes an intuitive interface that displays model outputs alongside the source text, allowing users to inspect, correct, and iteratively improve performance. MetaBeeAI produces an auditable, machine-readable record of prompts, configuration settings, and expert annotations, supporting reliable replication and continual refinement. We evaluated the pipeline on 924 research papers, extracting information on bee species, pesticides, exposure methodologies, and other environmental factors. Results highlight the value of expert-in-the-loop validation for prompt optimisation and show that MetaBeeAI can handle heterogeneous experimental designs and biological contexts. MetaBeeAI provides a general framework for structured knowledge extraction, enabling scalable, transparent, and reproducible evidence synthesis, and new approaches to accelerate discovery in the life sciences using AI.

Year:

2025

Nonstationary tidal processes, such as tidal rivers and storm surge, present challenges for analysis and prediction because their inherent nonstationarity encumbers the use of standard tidal analysis tools like harmonic analysis. Moreover, specialized approaches impose problem‐specific functional forms and rely on auxiliary data, limiting their applicability across different nonstationary tidal processes. Although Munk and Cartwright's tidal response method avoids these assumptions, its lack of automation has hindered broader application. Here, we develop a nonparametric, automated response‐based analysis procedure. Our approach embeds a class of neural networks capable of representing any arbitrary Volterra series—the mathematical basis of the response method—within the classic framework. Our model facilitates the inclusion of meteorological and other non‐tidal forcing. By explicitly accounting for nonstationarity, our method yields improved astronomical tidal estimates. We further devise a strategy to extract physical insights from the learned model, demonstrating its utility in studying the interaction and modulation of astronomical tides by external forcing. By taking a nonparametric approach, our framework enables the investigation of phenomena that heretofore could not be accounted for straightforwardly, as illustrated by several case studies on tide–surge interaction, riverine tides, and storm surge. These applications, and more, can be replicated with just three lines of code using the open‐source Python package, RTide.

Year:

2025

The accuracy of global tidal models degrades significantly in coastal and estuarine regions. These models are important for correcting measurements from satellite altimetry and are used in numerous scientific and engineering applications. The new Surface Water Ocean Topography (SWOT) mission is providing measurements at unprecedented horizontal resolution in these regions. These data present both the opportunity and the necessity to quantify and correct the spatial variability in the model inaccuracies specific to these regions. We develop a variational Bayesian framework for tidal harmonic analysis which can be applied to SWOT, and is especially useful for exploting the data from the Cal/Val phase. The approach demonstrates superior robustness to different types of noise contamination in comparison to conventional least‐squares approaches while providing full uncertainty estimation. By imposing a spatially coherent inductive bias on the model, we achieve superior harmonic constituent inference from temporally sparse but spatially dense data. Bayesian uncertainty estimation gives rise to statistical methods for outlier removal and constituent selection. Using our approach, we estimate a lower bound for the residual tidal variability for two SWOT Cal/Val passes (003 and 016) around the European Shelf to be on average. We also show similar estimates cannot be produced using standard least‐squares approaches. Tide gauge validation in the same region confirms the superiority of our empirical approach in coastal environments. Empirical corrections for the SWOT data products are provided alongside an open‐source Python package, VTide.

Year:

2025

Sequential Bayesian inference can be used for continual learning to prevent catastrophic forgetting of past tasks and provide an informative prior when learning new tasks. We revisit sequential Bayesian inference and assess whether using the previous task’s posterior as a prior for a new task can prevent catastrophic forgetting in Bayesian neural networks. Our first contribution is to perform sequential Bayesian inference using Hamiltonian Monte Carlo. We propagate the posterior as a prior for new tasks by approximating the posterior via fitting a density estimator on Hamiltonian Monte Carlo samples. We find that this approach fails to prevent catastrophic forgetting, demonstrating the difficulty in performing sequential Bayesian inference in neural networks. From there, we study simple analytical examples of sequential Bayesian inference and CL and highlight the issue of model misspecification, which can lead to sub-optimal continual learning performance despite exact inference. Furthermore, we discuss how task data imbalances can cause forgetting. From these limitations, we argue that we need probabilistic models of the continual learning generative process rather than relying on sequential Bayesian inference over Bayesian neural network weights. Our final contribution is to propose a simple baseline called Prototypical Bayesian Continual Learning, which is competitive with the best performing Bayesian continual learning methods on class incremental continual learning computer vision benchmarks.

Year:

2023

Collaborators (11)

Harry Siviter

Dr at School of Biological Sciences

University of Bristol

UNITED KINGDOM
View details

Héloïse F Stevance

University of Oxford

UNITED KINGDOM
View details

Andrew K. Jones

Senior Lecturer in Molecular Biology and Genomics

Oxford Brookes University

UNITED KINGDOM
View details

Lars Straub

Lecturer

University of Bern

SWITZERLAND
View details

Rachel Parkinson

Lecturer (Assistant Professor)

Queen Mary University of London

UNITED KINGDOM
View details

Ben C. Lambert

University of Oxford

UNITED KINGDOM
View details

Fabio Sgolastra

Research Entomologist

ALMA MATER STUDIORUM UNIVERSITY OF BOLOGNA

ITALY
View details

Edward Straw

Trinity College Dublin

IRELAND
View details

Dara Anne Stanley

Lecturer in Applied Entomology

Dublin City University

IRELAND
View details

Lina Herbertsson

Lund University

SWEDEN
View details

Chris Bass

Professor of Applied Entomology

University of Exeter Cornwall Campus

UNITED KINGDOM
View details