Mathematical AI for Decision and Discovery
Henry Moss speaking at a MARS event

I am Henry Moss, an Associate Professor (UK Reader) in Mathematics and AI in the School of Mathematical Sciences at Lancaster University, where I hold a UKRI Future Leaders Fellowship and lead the Mathematical AI for Decision and Discovery (MADD) group. I am part of MARS, Lancaster's centre for the mathematics of AI in real-world systems.

In MADD, we work with industry partners to remove innovation bottlenecks in areas such as drug discovery and sustainable engineering, where experiments are technically and financially costly. Much of our current work develops Gaussian process flows: a general mechanism for incorporating the full richness of real-world knowledge into machine learning. We also work on Bayesian optimisation, active learning and experimental design.

Industry collaborators from: Secondmind · Microsoft Research · Amazon · AstraZeneca · Boeing · BigHat BioSciences · Genentech · Evonik Industries · MediaTek Research · Reaction Engines · Quantum Motion · UK Atomic Energy Authority · Monumo · Instadeep
Academic collaborators from: Cambridge · Oxford · Imperial College London · University College London · King's College London · Bristol · Edinburgh · Birmingham · ETH Zürich · EPFL · Tübingen · Bern · Aarhus · Aalto · Ghent · Copenhagen · Karolinska Institutet · IST Austria · Stanford · Cornell · Pennsylvania · Washington State · British Columbia · New South Wales · Western Australia · Melbourne

Before Lancaster I was an Assistant Research Professor in DAMTP at Cambridge, a research scientist at Secondmind, and a machine learning consultant to AstraZeneca. In 2021, I graduated from Lancaster's STOR-i with a PhD in statistical machine learning, after reading mathematics at Emmanuel College, Cambridge.

You can reach me at .

Available positions
  • Postdoctoral positions. Four-year posts with a generous travel budget.
  • A visiting researcher fund for PhD students, postdocs and other researchers in this area who would like to spend time in the group.
Interested? .
Team
Henry Moss Henry Moss Reader in Mathematics and AILancaster UniversityProbabilistic machine learning for scientific discovery
Rui-Yang Zhang Rui-Yang Zhang PhD studentLancaster UniversityAI-driven drifter placement (STOR-i)
Thomas Cowperthwaite Thomas Cowperthwaite PhD studentUniversity of CambridgeAutomatic Bayesian climate science: discovering governing equations (AI4ER)
Gabriel Diaz-Aylwin Gabriel Diaz-Aylwin PhD studentLancaster UniversityML-driven design for fusion reactors (UKAEA)
Abiel Talwar Abiel Talwar PhD studentLancaster UniversityGuiding scientific discovery with generative AI (G Research)
Michael Dodds Michael Dodds PhD studentUniversity of CambridgeML-guided design for high-throughput optimisation of biologics (AstraZeneca)
Dawid Lipinski Dawid Lipinski PhD studentLancaster UniversityEquivariant generative models for materials microstructures
Sam Willis Sam Willis PhD studentUniversity of CambridgeGenAI for aerospace engineering
Ranvir Narang Ranvir Narang PhD studentLancaster UniversityGaussian process flows for physics-informed modelling
Thomas Christie Thomas Christie Visiting scholarUniversity of TübingenGuidance in discrete diffusion models; benchmarking high-dimensional BO
Colin Doumont Colin Doumont Visiting scholarUniversity of TübingenKernels and linear models for high-dimensional Bayesian optimisation
Featured Research: Gaussian Process Flows
A Gaussian process flow: white noise gradually becoming smooth sample paths that pass through a set of fixed observations, shown in red.

Sampling a Gaussian process by flowing white noise into a predictive distribution. Because the samples appear gradually, we can steer them as they form. This gives a general mechanism for incorporating the full richness of real-world knowledge as conditioning information, opening a new frontier for the probabilistic modelling of real-world problems.

This programme of research is led jointly with Lachlan Astfalck.

Conditioning Gaussian Processes on Almost Anything Feature Space Adaptation for Effortless Gaussian Process Flows LatentFlow: A General Framework for Conditioning Stochastic Processes
Publications

Also on Google Scholar.

Preprints
A Critical look at Power Sampling when Test-time Sharpening LLMs A Critical look at Power Sampling when Test-time Sharpening LLMsTalwar, Nemeth, Moss PDF
Feature Space Adaptation for Effortless Gaussian Process Flows Feature Space Adaptation for Effortless Gaussian Process FlowsCowperthwaite, Sharrock, Astfalck, Moss PDFarXiv
Example-driven Parametrisations for Bayesian Shape Optimisation Example-driven Parametrisations for Bayesian Shape OptimisationDiaz-Aylwin, Neighbour, Talwar, Zhang, Moss PDFarXiv
Bayesian Optimisation under State-Preservation Constraints Bayesian Optimisation under State-Preservation ConstraintsDiaz-Aylwin, Gopakuar, Myatra, Moulton, Zanisi, Leslie, Moss PDFarXiv
LatentFlow: A General Framework for Conditioning Stochastic Processes LatentFlow: A General Framework for Conditioning Stochastic ProcessesSharrock, Astfalck, Moss PDFarXiv
Sample-Efficient Optimisation over the Outputs of Generative Models Sample-Efficient Optimisation over the Outputs of Generative ModelsWillis, Duckworth, Simons, Kalisz,..., Moss, Bodin PDFarXiv
RF-Squad: A Radiofrequency Simulator for Quantum Dot Arrays RF-Squad: A Radiofrequency Simulator for Quantum Dot ArraysMurphy, Brlec, Oakes, Peri, Sirringhaus, Moss, Gonzalez-Zalba, Wise PDFarXiv
Don't Get Your Kroneckers in a Twist: Gaussian Processes over Sub-grids Don't Get Your Kroneckers in a Twist: Gaussian Processes over Sub-gridsHøjlund, Lykke-Møller, Moss, Christiansen PDFarXiv
Dynamic Gaussian Processes and the Vanilla-SPDE Exchange Dynamic Gaussian Processes and the Vanilla-SPDE ExchangeZhang, Astfalck, Cripps, Leslie, Moss PDFarXiv
2026
Conditioning Gaussian Processes on Almost Anything Conditioning Gaussian Processes on Almost AnythingMoss, Astfalck, Cowperthwaite, Doumont, Willis, Nemeth, Hennig, Zammit-Mangion NeurIPS 2026 PDFarXiv
Search at the Cost of Sampling: Nearly-Instant Latent Space Bayesian Optimization Search at the Cost of Sampling: Nearly-Instant Latent Space Bayesian OptimizationFan, Doumont, Kalisz, Duckworth, Gardner, Moss, Pleiss NeurIPS 2026 PDFarXiv
Active Learning for Gaussian Process Regression Under Self-Induced Boltzmann Weights Active Learning for Gaussian Process Regression Under Self-Induced Boltzmann WeightsQing, Moss, Sachs NeurIPS 2026 PDFarXiv
We Still Don't Understand High-Dimensional Bayesian Optimisation We Still Don't Understand High-Dimensional Bayesian OptimisationDoumont, Fan, Maus, Gardner, Moss, Pleiss AISTATS 2026★ Best Paper Award — top 3 of 2,500 submissions PDFarXiv
BALLAST: Spatio-Temporal Active Learning of Time-Dependent Vector Fields using Lagrangian Observers BALLAST: Spatio-Temporal Active Learning of Time-Dependent Vector Fields using Lagrangian ObserversZhang, Astfalck, Cripps, Leslie, Moss ICML 2026 PDFarXiv
Modelling Sub-kilometre Surface Wind with Gaussian Processes and Neural Networks Modelling Sub-kilometre Surface Wind with Gaussian Processes and Neural NetworksZanetta, et al., Moss Artificial Intelligence for the Earth Systems, 2026 arXiv
Replacing Tunable Parameters in Weather and Climate Models with State-Dependent Functions using Reinforcement Learning Replacing Tunable Parameters in Weather and Climate Models with State-Dependent Functions using Reinforcement LearningNath, Schemm, Moss, Haynes, Shuckburgh, Webb Journal of Advances in Modeling Earth Systems, 2026 PDFarXiv
Advancing the Classification of Supraglacial Lake Winter Behaviours on the Greenland Ice Sheet Advancing the Classification of Supraglacial Lake Winter Behaviours on the Greenland Ice SheetOtto, Miles, Leeson, Maddalena, McMillan, Moss Journal of Glaciology, 2026
Conditioning Stochastic Processes using Guided Latent Dynamics Conditioning Stochastic Processes using Guided Latent DynamicsSharrock, Astfalck, Moss NeurIPS 2026 Workshop on AI for Stochastic Dynamics PDF
Scalable Gaussian Process Flows Scalable Gaussian Process FlowsCowperthwaite, Sharrock, Astfalck, Moss NeurIPS 2026 Workshop on AI for Stochastic Dynamics PDF
How Reliable are Intermediate Rewards for SMC-based Guidance in Masked Discrete Diffusion Models? How Reliable are Intermediate Rewards for SMC-based Guidance in Masked Discrete Diffusion Models?Christie, Ek, Hennig, Moss NeurIPS 2026 Workshop on Principles of Generative Modelling PDF
C4-Equivariant Flow Matching on Anisotropic Power-Diagram Graphs for Microstructure Generation C4-Equivariant Flow Matching on Anisotropic Power-Diagram Graphs for Microstructure GenerationLipinski, Buze, Moss NeurIPS 2026 Workshop on Representations for the Physical Sciences PDF
How to Spend Your Oracle Budget: Practical Guidance for Protein Structure Foundation Models How to Spend Your Oracle Budget: Practical Guidance for Protein Structure Foundation ModelsKalisz, Simons, Sinkovics, Ghenassia, Surana, Moss, Duckworth ICML 2026 SPIGM Workshop arXiv
2025
Omnipresent Yet Overlooked: Heat Kernels in Combinatorial Bayesian Optimization Omnipresent Yet Overlooked: Heat Kernels in Combinatorial Bayesian OptimizationDoumont, Picheny, Borovitskiy, Moss NeurIPS 2025 PDFarXiv
Return of the Latent Space COWBOYS Return of the Latent Space COWBOYSMoss, Ober, Diethe ICML 2025★ Spotlight PDFarXiv
Integration-free Kernels for Equivariant Gaussian Process Modelling Integration-free Kernels for Equivariant Gaussian Process ModellingSteinert, Ginsbourger, Moss ICML 2025 PDF
Linear Combinations of Latents in Diffusion Models: Interpolation and Beyond Linear Combinations of Latents in Diffusion Models: Interpolation and BeyondBodin, Stere, Margineantu, Ek, Moss ICLR 2025 PDFarXiv
Reflective Error: A Metric for Assessing Predictive Performance at Extreme Events Reflective Error: A Metric for Assessing Predictive Performance at Extreme EventsRouse, Moss, Hosking, McRobie, Shuckburgh Environmental Data Science, 2025
GPGreen: Learning Linear Operators with Gaussian Processes GPGreen: Learning Linear Operators with Gaussian ProcessesCowperthwaite, Moss EurIPS 2025 Workshop on Differentiable Systems and Scientific Machine Learning PDF
Enabling Efficient Experimental Design in the Context of High-Dimensional Generative Models Enabling Efficient Experimental Design in the Context of High-Dimensional Generative ModelsWillis, Oldroyd, Fozard, Ek, Moss EurIPS 2025 Workshop on Machine Learning for Simulations in Biology and Chemistry PDF
FedRAIN-Lite: Federated Reinforcement Algorithms for Improving Idealised Numerical Weather and Climate Models FedRAIN-Lite: Federated Reinforcement Algorithms for Improving Idealised Numerical Weather and Climate ModelsNath, Moss, Shuckburgh, Webb Eurips 2025 Workshop on AI for Climate and Conservation PDFarXiv
2024
Bayesian Optimisation for Additive Screening and Yield Improvements — Beyond One-Hot Encoding Bayesian Optimisation for Additive Screening and Yield Improvements — Beyond One-Hot EncodingRanković, Griffiths, Moss, Schwaller Digital Discovery, 2024
Big Batch Bayesian Active Learning by Considering Predictive Probabilities Big Batch Bayesian Active Learning by Considering Predictive ProbabilitiesOber, Power, Diethe, Moss NeurIPS 2024 Workshop on Bayesian Deep Learning and Uncertainty Quantification PDFarXiv
Trieste: Efficiently Exploring the Depths of Black-box Functions with TensorFlow Trieste: Efficiently Exploring the Depths of Black-box Functions with TensorFlowPicheny, Berkeley, Moss, Stojić, Granta, Ober, Artemev, et al. NeurIPS 2024 Workshop on Bayesian Deep Learning and Uncertainty Quantification PDFarXivCode
Integration-free kernels for equivariant Gaussian fields with application in dipole moment prediction Integration-free kernels for equivariant Gaussian fields with application in dipole moment predictionSteinert, Ginsbourger, Moss NeurIPS 2024 Workshop on Bayesian Deep Learning and Uncertainty Quantification PDF
BALLAST: Bayesian Active Learning with Look-ahead Amendment for Sea-drifter Trajectories BALLAST: Bayesian Active Learning with Look-ahead Amendment for Sea-drifter TrajectoriesZhang, Moss, Astfalck, Cripps, Leslie NeurIPS 2024 Workshop on Bayesian Deep Learning and Uncertainty Quantification PDF
RAIN: Reinforcement Algorithms for Improving Numerical Weather and Climate Models RAIN: Reinforcement Algorithms for Improving Numerical Weather and Climate ModelsNath, Moss, Shuckburgh, Webb NeurIPS 2024 Workshop on Tackling Climate Change with Machine Learning PDFarXiv
MONGOOSE: Path-wise Smooth Bayesian Optimisation via Meta-learning MONGOOSE: Path-wise Smooth Bayesian Optimisation via Meta-learningYang, Aitchison, Moss ICML 2024 SPIGM Workshop PDFarXiv
2023
GAUCHE: A Library for Gaussian Processes in Chemistry GAUCHE: A Library for Gaussian Processes in ChemistryGriffiths, Klarner, Moss, Ravuri, Truong, Ranković, Schwaller, et al. NeurIPS 2023 PDFarXivCode
Inducing Point Allocation for Sparse Gaussian Processes in High-throughput Bayesian Optimisation Loops Inducing Point Allocation for Sparse Gaussian Processes in High-throughput Bayesian Optimisation LoopsMoss, Ober, Picheny AISTATS 2023★ Oral PDFarXiv
PF²ES: Parallel Feasible Pareto Frontier Entropy Search for Multi-Objective Bayesian Optimization under Unknown Constraints PF²ES: Parallel Feasible Pareto Frontier Entropy Search for Multi-Objective Bayesian Optimization under Unknown ConstraintsQing, Moss, Couckuyt AISTATS 2023 PDFarXiv
2022
Bayesian Quantile and Expectile Optimisation Bayesian Quantile and Expectile OptimisationPicheny, Moss, Durrande, Torossian UAI 2022 PDFarXiv
Data-driven Discovery of Molecular Photoswitches with Multioutput Gaussian Processes Data-driven Discovery of Molecular Photoswitches with Multioutput Gaussian ProcessesGriffiths, Thawani, Jamasb, Moss, Bourached, Jones, McCorkindale, Aldrick Chemical Science, 2022 PDF
Fantasizing with Dual GPs in Bayesian Optimization and Active Learning Fantasizing with Dual GPs in Bayesian Optimization and Active LearningChang, Verma, John, Moss, Picheny, Solin NeurIPS 2022 GP Workshop PDFarXiv
A Penalisation Method for Batch Multi-objective Bayesian Optimisation with Application in Heat-exchanger Design A Penalisation Method for Batch Multi-objective Bayesian Optimisation with Application in Heat-exchanger DesignPaleyes, Moss, Picheny ICML 2022 RealML Workshop PDFarXiv
Information-theoretic Inducing Point Placement for High-throughput Bayesian Optimisation Information-theoretic Inducing Point Placement for High-throughput Bayesian OptimisationMoss, Ober,Picheny ICML 2022 RealML Workshop PDFarXiv
2021 and earlier (PhD)
GIBBON: General-purpose Information-based Bayesian OptimisatioN GIBBON: General-purpose Information-based Bayesian OptimisatioNMoss, Leslie, González, Rayson Journal of Machine Learning Research, 2021 PDFarXiv
Scalable Thompson Sampling using Sparse Gaussian Process Models Scalable Thompson Sampling using Sparse Gaussian Process ModelsVakili, Moss, Artemev, Dutordoir, Picheny NeurIPS 2021 PDFarXiv
BOSS: Bayesian Optimisation over String Spaces BOSS: Bayesian Optimisation over String SpacesMoss, Beck, Leslie, González, Rayson NeurIPS 2020★ Spotlight PDFarXiv
MUMBO: MUlti-task Max-value Bayesian Optimisation MUMBO: MUlti-task Max-value Bayesian OptimisationMoss, Leslie, Rayson ECML-PKDD 2020 PDFarXiv
BOFFIN TTS: Few-shot Speaker Adaptation by Bayesian Optimisation BOFFIN TTS: Few-shot Speaker Adaptation by Bayesian OptimisationMoss, Aggarwal, Prateek, González, Barra-Chicote ICASSP 2020 PDFarXiv
Gaussian Process Molecule Property Prediction with FlowMO Gaussian Process Molecule Property Prediction with FlowMOMoss, Griffiths NeurIPS 2020 ML4Molecules Workshop★ Selected talk PDFarXiv
BOSH: Bayesian Optimisation by Sampling Hierarchically BOSH: Bayesian Optimisation by Sampling HierarchicallyMoss, Leslie, Rayson ICML 2020 RealML Workshop PDFarXiv
FIESTA: Fast IdEntification of State-of-The-Art Models Using Adaptive Bandit Algorithms FIESTA: Fast IdEntification of State-of-The-Art Models Using Adaptive Bandit AlgorithmsMoss, Moore, Leslie, Rayson ACL 2019 PDFarXiv
Using J-K-fold Cross Validation to Reduce Variance when Tuning NLP Models Using J-K-fold Cross Validation to Reduce Variance when Tuning NLP ModelsMoss, Leslie, Rayson COLING 2018★ Area chair favourite PDFarXiv