Seminars

Time: 13.00-14.00

Seminar location: A2.05B
There will be an informal sandwich lunch outside D2.02 at 12.30.

To join this meeting online click here.Link opens in a new window

Title: Robust error-controlled materials simulations

Abstract: Systematic first-principle simulations are nowadays a key component when developing novel materials. Usually the resulting simulation data is not directly used to drive the search, but instead employed to train a considerable cheaper statistical surrogate. In this setting of potentially millions of simulations as well as these multiple layers of approximations (physical, numerical, statistical) obtaining robust computational workflows and tracking simulation errors remains challenging.

In this talk I will report on progress along two axes of research to tackle these challenges. The first concerns the development of robust numerical algorithms for density-functional theory (DFT) --- the most widely employed family of first-principle models in the field. The focus of the development here is to obtain black-box methods that are able to automatically adapt to the physics of the simulated system.

Secondly, I will discuss first results on employing multi-task statistical surrogate models, a surrogatisation technique, which enables the use of data of heterogeneous quality when training a single surrogate. By combining materials simulation approaches of different cost/accuracy balances this not only unlocks computational savings to generate training data, but also allows to opportunistically exploit heterogeneous databases of already existing simulation data.

In both efforts software has played a key role to provide an accessible platform fostering such interdisciplinary developments. In our work we develop and extend the density-functional toolkit (DFTK), a Julia-based DFT code, suitable to mathematical research (only 7500 lines of code), but at the same time integrated into standard tools for materials discovery.

Time: 13.00-14.00

Seminar location: A2.05B
There will be an informal sandwich lunch outside D2.02 at 12.30.

To join this meeting online click here.Link opens in a new window

Title: Structural Identifiability Analysis: An Important Tool in Systems Modelling

Abstract: For many systems (certainly those in biology, medicine and pharmacology) the mathematical models that are generated invariably include state variables that cannot be directly measured and associated model parameters, many of which may be unknown, and which also cannot be measured. For such systems there is also often limited access for inputs or perturbations. These limitations can cause immense problems when investigating the existence of hidden pathways or attempting to estimate unknown parameters and this can severely hinder model validation. It is therefore highly desirable to have a formal approach to determine what additional inputs and/or measurements are necessary in order to reduce or remove these limitations and permit the derivation of models that can be used for practical purposes with greater confidence. 

Structural identifiability arises in the inverse problem of inferring from the known, or assumed, properties of a system a suitable model structure and estimates for the corresponding rate constants and other model parameters. Structural identifiability analysis considers the uniqueness of the unknown model parameters from the input-output structure corresponding to proposed experiments to collect data for parameter estimation (under an assumption of the availability of continuous, noise-free observations). This is an important, but often overlooked, theoretical prerequisite to experiment design, system identification and parameter estimation, since estimates for unidentifiable parameters are effectively meaningless. If parameter estimates are to be used to inform about intervention, inhibition or control strategies, or other critical decisions, then it is essential that the parameters be uniquely identifiable.  

Numerous techniques for performing a structural identifiability analysis on linear parametric models exist and this is a well-understood topic. In comparison, there are relatively few techniques available for nonlinear systems (the Taylor series approach, similarity transformation-based approaches, differential algebra techniques and the more recent observable normal form approach and symmetries approaches) and significant (symbolic) computational problems can arise, even for relatively simple models in applying these techniques. 

In this talk an introduction to structural identifiability analysis will be provided demonstrating the application of the techniques available to both linear and nonlinear parameterised systems. A brief overview of current research in this field will also be provided. 

Bio: Professor Michael Chappell | School of Engineering | University of Warwick

Time: 13.00-14.00

Seminar location: A2.05B
There will be an informal sandwich lunch outside D2.02 at 12.30.

To join this meeting online click here.Link opens in a new window

Title: Atomistic Simulations of Dislocation – Precipitate Interactions in Ni-based Superalloys

Abstract: The interaction between dislocations and precipitates is one of the archetypical hardening mechanisms in alloys and plays together with solid solution strengthening a critical role for the high (creep) strength of superalloys. Precipitation hardening in superalloys has been intensively studied, e.g., through TEM observations of interrupted creep tests, and more recently by in-situ micromechanical tests. Analytical and numerical models can successfully describe many aspects of strengthening on the meso- and macroscale. However, such models as discrete dislocation dynamics (DDD) simulations require parameters, which depend on the atomic-scale details of the dislocation – precipitate interactions. Atomistic simulations can in principle provide such information but are currently severely limited by the lack of accurate atomic interaction potentials for technologically relevant, multi-component alloys and by the difficulties to include diffusive processes. Therefore, there are currently still relatively few atomistic simulations of dislocation-precipitate interactions in superalloys.

Here we present an overview of our atomistic simulations in the Ni-Al-(Re) system as model for γ/γ’ strengthened alloys. We show that while parameters like the cutting-stress for dislocations to enter γ’ precipitates can be obtained from idealized geometries, the details of the γ/γ’ interface structure, the precipitate morphology and arrangement can severely influence the dislocation-precipitate interactions. In particular the curvature of the γ/γ’ interface can affect the misfit dislocation network, as demonstrated using experimentally obtained γ/γ’ interface morphologies. The local interface orientation not only alters the misfit dislocation core structure but can also facilitate the formation of ⟨100⟩ dislocations at the interface. Furthermore, the spatial arrangement and size-distribution of spherical γ’ precipitates, e.g., in disk-alloys, can lead to synergistic effects that are not present in the typical models of precipitate strengthening based on the interaction of straight dislocations with a regular array of uniform precipitates. Certain Ni-base superalloys furthermore form γ precipitates inside the cuboidal γ’ phase. Our simulations suggest that the misfit stresses caused by the γ precipitates reduce the yield stress of γ’ cubes subjected to nanomechanical compression tests. The situation is, however, different when the deformation is not controlled by the nucleation of dislocations, e.g., when the γ’ cubes are embedded in a dislocation-containing γ matrix. In this case, the γ precipitates lead to an additional hardening that is also observed experimentally.

Time: 13.00-14.00

Seminar location: A2.05B
There will be an informal sandwich lunch outside D2.02 at 12.30.

To join this meeting online click here.Link opens in a new window

Title: Predicting the phase stability of BaZrS3 using a range of approaches: from harmonic lattice dynamics to the neuroevolution-potential framework

Location: A2.05B (there will be an informal sandwich lunch outside D2.02 at 12.30)

Time: 13.00-14.00

To join this meeting online click here.

Title: Theoretical Study on Local Structures in Fe-based Amorphous Alloys

Abstract: Local atomic arrangements, such as short-range order and medium-range order, are crucial in determining the physical properties of amorphous materials. Direct experimental observation of these microscopic structures is challenging; nevertheless, computational approaches such as molecular dynamics simulations afford valuable insights.

This study investigates Fe–Si–B amorphous alloys, recognised for their excellent soft magnetic properties, making them promising candidates for high-efficiency motor core materials. However, their challenging manufacturability has limited their widespread adoption. Addressing this issue requires a deeper understanding of the origins of their mechanical properties.

In our study, melt-quench simulations were conducted using a machine learning potential based on the Gaussian approximation potential (GAP). The radial distribution functions obtained from these simulations closely matched the experimental data. The results revealed that B-centred clusters predominantly exhibit anti-prism and trigonal prism structures, while the Si-centred clusters are characterised by icosahedral and related geometries.

These cluster structures formed rapidly near the glass transition temperature during quenching and subsequently coalesced into networks. The modes of connection between cluster pairs—vertex, edge, face, and multipoint sharing—were found to depend on the cluster type. The hierarchical ordering observed in these alloys should play a crucial role in understanding their mechanical behaviour.

This study highlights the potential of GAP for uncovering the structural properties of amorphous alloys, offering a pathway to improve their manufacturability and broader applicability.

Bio:

Background

· 2020–Present: Associate Professor (Next Generation Tatara Co-Creation Centre, Shimane University)

· 2009–2020: Project Researcher (University of Tokyo, Osaka University, and others); Assistant Professor (Tokyo University of Science, Tokyo Metropolitan University)

· March 2009: PhD in Science (Ochanomizu University, Tokyo, Japan)

Research Field

· Amorphous alloys: Investigation of structural and mechanical properties using machine-learning molecular dynamics simulations

· Thermoelectric materials: Calculation of electronic and thermoelectric properties based on first-principles calculations.

Location: A2.05B (there will be an informal sandwich lunch outside D2.02 at 12.30)

Time: 13.00-14.00

To join this meeting online click here. Link opens in a new windowLink opens in a new window

Title: Towards Deep Learning Surrogate Modelling with Uncertainty Quantification in Mechanics
Jakub Lengiewicz, Luxembourg Institute of Science and Technology

In this talk, I will present recent advances in using deep learning to build fast and reliable surrogate models for computational mechanics. These methods aim to accelerate predictions of large-deformation responses in solids—scenarios where traditional finite element analysis can be prohibitively expensive—while also providing robust uncertainty estimates. I will discuss a variety of neural architectures, including convolutional and graph-based U-Nets, as well as attention-based models, that can accurately learn non-linear mechanical behaviour directly from simulation data. In addition, I will show how Bayesian techniques enable probabilistic modelling, offering meaningful confidence measures in complex predictive tasks. Together, these approaches pave the way toward efficient, data-driven computational modelling that is both fast and reliable, with potential applications in engineering design, materials science, and beyond.

Bio: Jakub Lengiewicz is a Senior Research and Technology Scientist at the Luxembourg Institute of Science and Technology and is affiliated with the Institute of Fundamental Technological Research of the Polish Academy of Sciences. He holds a background in Computer Science, a PhD in Mechanics, and a habilitation in Information and Communication Technologies. His research expertise spans computational methods in mechanics and robotics, with contributions in finite element techniques for contact mechanics and tribology, as well as distributed algorithms for modular robotic systems. Since his Marie Skłodowska-Curie Postdoctoral Fellowship at the University of Luxembourg (initiated in 2019), he has focused on deep learning surrogate modelling in mechanics, a pursuit he continues at the Luxembourg Institute of Science and Technology

Location: A2.05B (there will be an informal sandwich lunch outside D2.02 at 12.30)

Time: 13.00-14.00

To join this meeting online click here.

Link opens in a new windLink opens in a new windowLink opens in a new windowLink opens in a new window

Location: A2.05B (there will be an informal sandwich lunch outside D2.02 at 12.30)

Time: 13.00-14.00

To join this meeting online click here.

Link opens in a new windLink opens in a new windowLink opens in a new windowLink opens in a new windowLink opens in a new window

Title: Strategies for including non-adiabatic effects in ab initio materials modeling

Abstract:

The advancement of ab initio methods that include non-adiabatic effects is an active research area with important consequences for materials modelling and surface chemistry. Non-adiabatic effects originate from transitions between adiabatic electronic eigenstates. Such transitions are induced by the nuclear kinetic energy operator or external electromagnetic fields and other external sources. Since the nuclear kinetic energy operator contains a small numerical prefactor, the inverse nuclear mass, it is tempting to treat it perturbatively. This is the foundation of many approaches to electron-phonon systems but comes with limitations. Quantum chemistry methods on the other hand are non-perturbative, yet it is challenging to make them practical for extended systems.

To develop a universal perturbation theory of non-adiabatic effects with controlled accuracy in terms of the electron-to-nucleus mass ratio, it is necessary to treat the nuclear kinetic energy operator as a singular perturbation. Because singular perturbations fundamentally alter a problem when they are removed, e.g. for the purpose of defining a zeroth-order Hamiltonian, they cannot be handled by standard perturbation theories. I will overview novel strategies for incorporating non-adiabatic effects in ab initio calculations, including exact-factorization-based density functional theory [1] and adiabatic perturbation theory [2] as well as our progress on implementions.

[1] R. Requist, C. R. Proetto, E. K. U. Gross, Phys. Rev. B 99, 165136, 2019.
[2] R. Requist, Phys. Rev. B 111, 024311, 2025.

Location: A2.05B (there will be an informal sandwich lunch outside D2.02 at 12.30)

Time: 13.00-14.00

To join this meeting online click here.

Link opens in a new windoLink opens in a new windowLink opens in a new window

Modelling severity of emerging epidemics via state-space models and the lifebelt particle filter.

The inference of the severity (measured as probability of death) of an emerging epidemic is a non-trivial statistical challenge. Information on the most severe events is often collected in terms of time series of the counts of severely symptomatic cases and deaths per day/week; no data are usually available on the number of patients who recover from the disease or on timing between consecutive events. Cumulative-based counts estimators are well known to be biased but they are still considered the gold standard, as the lack of suitable data seems to prevent the use alternative methods.

In this talk we approach this problem from a state-space model (SSM) perspective, properly accounting for all the unknown underlying processes. Particle-filtering methods can be used to drive Bayesian inference of the severity parameters of interest, but they are prone to fail in this context given the low counts and discrete system. We propose the Lifebelt Particle Filter (LBPF), a new particle method that is robust to particle collapse thanks to the use of deterministic mixtures in the importance distribution and a conservative resampling scheme

Location: A2.05B (there will be an informal sandwich lunch outside D2.02 at 12.30)

Time: 13.00-14.00

To join this meeting online click here.

Link opens in a new windowLink opens in a new windowLink opens in a new window

Title: First principles simulations of electron-phonon coupling and thermoelectric transport in PbTe

Location: A2.05B (there will be an informal sandwich lunch outside D2.02 at 12.30)

Time: 13.00-14.00

To join this meeting online click here.

Bio:

Location: B2.02 Science Concourse (there will be an informal sandwich lunch outside B2.02 at 12.30)

Time: 13.00-14.00

To join this meeting online click here.

Title: Adaptive Particle Refinement in Smoothed Particle Hydodynamics.
Speaker: Rebecca Nealon

Abstract: Adaptive mesh refinement is commonly used in grid based simulations to locally increase resolution and improve computational efficiency. However, such a method is essentially non-existent in compressible smoothed particle hydrodynamics (SPH), commonly used for astrophysical applications. In this talk I will describe our implementation of a novel adaptive particle refinement scheme for SPH. I will describe particular applications of this method and demonstrate its efficiency, accuracy, particularly useful applications and extensions.

Bio: Rebecca is a Stephen Hawking Fellow and Assistant Professor in the Astronomy and Astrophysics group at the University of Warwick. Her research focuses on warped and misaligned accretion discs and planet formation. In particular she considers protoplanetary discs (accretion discs around young stars) as well as black hole discs.

In her work she uses high resolution three-dimensional numerical simulations to model these systems. To see some of these simulations and learn more about her work, please do refer to her website or youtube page.

B2.02

Location: B2.02 Science Concourse (there will be an informal sandwich lunch outside D2.02 12.30pm)

Time: 13.00-14.00

To join this meeting online click here

Location: B2.02 Science Concourse (there will be an informal sandwich lunch outside D2.02 12.30pm)

Time: 13.00-14.00

Time: 13.00-14.00

Seminar location: B2.02 (There will be an informal sandwich lunch outside D2.02 at 12.30)

Time: 13.00-14.00

Seminar location: B2.02 (There will be an informal sandwich lunch outside D2.02 at 12.30)

Time: 13.00-14.00

Seminar location: B2.02

There will be an informal sandwich lunch outside D2.02 at 12.30.

Time: 13.00-14.00

Seminar location: B2.02

There will be an informal sandwich lunch outside D2.02 at 12.30.

Location: Lecture Theatre 0.04 IMC

Networking Lunch: The Recharge Room, next to Lecture Theatre 004, from 12:30pm - 1pm.

Title: TBC

Abstract: TBC

Bio: TBC

Location:  Lecture Theatre 0.04 IMC

Networking Lunch: The Recharge Room, next to Lecture Theatre 004, from 12:30pm - 1pm.

Title: TBC

Abstract: TBC

Bio: TBC

Bio: Dr Luke Davis is a theoretical physicist broadly working in the area of statistical physics, stochastic systems, soft matter, and theoretical biophysics. Currently, he is a (4-year) Flora Philip Fellow based at The School of Mathematics, Applied and Computational Mathematics, at the University of Edinburgh. His previous position was as an Isaac Newton Institute (INI) Postdoctoral Research Fellow at the University of Cambridge.

Current questions/topics that he is thinking and researching about:

• How to best steer continuous and discrete-state active matter to perform useful functions efficiently?
• Mathematical and computational exploration of the stochastic geometry of active matter.
• Exploring efficient and robust optimization of active matter simulations.
• How to adapt ideas and techniques from equilibrium thermodynamics to better understand the physics of active matter?
• Biophysics, in particular modelling assemblies of disordered proteins and biomolecular condensates that may also be out-of-equilibrium.

Location: Lecture Theatre 0.04 IMC

Networking Lunch: The Recharge Room, next to Lecture Theatre 004, from 12:30pm - 1pm.

Title: TBC

Abstract: TBC

Bio: The Bevan Research Group explores nanoscale materials and devices via “technology computer aided design” (TCAD) to develop next-generation energy, computing, and sensing technologies. The ultimate goal of this research is to drive the discovery of new technologies through “electronic design automation” (EDA). Through the application and development of advanced theoretical and machine learning methods, group members research materials problems limiting development of the aforementioned technological fields. Recent, ongoing, and previous research applications include: next-generation batteries, photo-electrolysis, supercapacitors, semiconductor devices, oxide electronics, molecular devices, CO2 reduction, electrocatalysis, and advanced materials synthesis/growth. These efforts are rooted in the exploration of materials from their fundamental governing principles (e.g., quantum), whereby material properties are tailored through atomic-scale and nano-scale modeling methods. Research is often conducted in close collaboration with experimental groups to enable the rapid discovery of new materials & devices for energy, electronic, and sensing applications.

Location:  Lecture Theatre 0.04 IMC

Networking Lunch: The Recharge Room, next to Lecture Theatre 004, from 12:30pm - 1pm.

Title: TBC

Abstract: TBC

Bio: Dr Nick Tawn is a Reader in Statistics at the University of Warwick.

His research interests have mostly focussed on developing scalable Monte Carlo methodology for use in complex Bayesian settings; in particular MCMC and SMC techniques. He also takes a keen interest in the Machine Learning/Data Science literature with a view to using these methods to complement and accelerate the inference process.

More specifically his research has focussed on MCMC methodology in settings where the target distribution exhibits multi-modality. My PhD thesis, completed in 2017, was titled "Towards Optimality of the Parallel Tempering Algorithm". Since completion of his thesis he has continued to work on similar problems whilst writing up the ideas from my thesis for publication.

Location:  Lecture Theatre 0.04 IMC

Networking Lunch: The Recharge Room, next to Lecture Theatre 004, from 12:30pm - 1pm.

Title: TBC

Abstract: TBC

Bio: Dr Emma Davis is a mathematical modeller whose research focuses on the dynamics of endemic and emerging infectious diseases. She uses mathematical models and statistical methods to investigate disease transmission, surveillance, and elimination strategies, with applications ranging from Neglected Tropical Diseases (NTDs) to the COVID-19 pandemic.

Her recent work includes developing models to support the detection of lymphatic filariasis resurgence in post-elimination settings, analysing the impact of vector control on mosquito-borne infections, and assessing the consequences of programme interruptions during the COVID-19 crisis on global NTD control.

Dr Davis teaches MA147: Mathematical Methods and Modelling 1 (for joint honours students) and is also committed to public engagement with science. She runs the YouTube channel “Epi with Emma”, which brings epidemiological modelling to a wider audience.

Her research has been published in leading journals including Nature Communications, The Lancet Global Health, and Philosophical Transactions of the Royal Society B. She has been recognised with multiple awards, including the Weldon Memorial Prize (2022) as part of SPI-M-O, the SPI-M-O Award for Modelling and Data Support (2022) from the UK Department of Health and Social Care, and the RAMP Early Career Investigation Award (2021) from the Royal Society.

Location:  Lecture Theatre 0.04 IMC

Networking Lunch: The Recharge Room, next to Lecture Theatre 004, from 12:30pm - 1pm.

Title: TBC

Abstract: TBC

Bio: Dr. Edwina Yeo is an applied mathematician whose research lies at the intersection of continuum mechanics and mathematical biology. She is currently an EPSRC National Fluid Dynamics (NFFDy) Fellow in the Department of Mathematics at University College London (UCL), where she develops continuum models to study aggregating fluid systems.

Before joining UCL, Dr. Yeo was an EPSRC Doctoral Prize Postdoctoral Researcher at the Mathematical Institute, University of Oxford. She earned her DPhil in Mathematics (2023) and an MSc in Mathematical Modelling and Scientific Computing (2018), both from the University of Oxford.

Her work combines rigorous mathematical analysis with practical applications, contributing to a deeper understanding of complex fluid dynamics and its role in biological and physical systems.

Location: Lecture Theatre 0.04 IMC

Networking Lunch: The Recharge Room, next to Lecture Theatre 004, from 12:30pm - 1pm.

Title: Synergy of Theory and Experimental NMR in Energy Material Research

Abstract: Nuclear magnetic resonance (NMR) spectroscopy provides a powerful tool for probing high-performance energy materials such as solid ion conductors. Probing different spin interactions enables insights into atomic dynamics across a range of time and length scales but requires computational simulation to analyse the spectral structure-property relationships. However, bridging the gap between the complexity of experimental samples and the simplifications and approximations inherent in computational model systems is challenging to tackle.

Our multi-scale ansatz starts with plane-wave density functional theory (DFT) to simulate NMR tensorial properties and their derived observables from first principles. DFT provides the high-quality reference NMR tensors for tensorial machine learning (ML) in order to predict NMR with comparable computational efficiency to long-timescale MD with machine learned interatomic potentials (MLIP). By combining MD simulations with ML-based NMR predictions, NMR-relevant dynamics over experimentally relevant timescales are directly simulated capturing the evolution of structure–property relationships with high accuracy. By the addition of the experimental postprocessing workflow, our approach opens the door to predictive in silico NMR experiments that reveal how local atomic environments govern macroscopic behaviour in complex materials. Finally, the simulation of quantum dynamics enables us to customise NMR experiments and increase their selectivity for electrochemical interfaces.

Bio: Simone Köcher studied chemistry at the Technical University Munich with a focus on theoretical chemistry and magnetic resonance. She conducted her PhD at IEK-9 Forschungszentrum Jülich and RWTH Aachen with Prof. Josef Granwehr in cooperation with Prof. Karsten Reuter at TU Munich simulating lithium ion battery materials and computing their spectroscopic as well as dynamic properties. After a PostDoc at TU Munich working on parallel eigensolvers (ELPA) and their implementation in electronic structure codes in collaboration with the Max Planck Computing and Data Facility (MPCDF), she joint Prof. Stefano Sanvito at Trinity College Dublin to study magnetic materials with first principles as well as machine learning methods. In 2022, she returned to the IEK-9, now IET-1, to head the new department of Theoretical Electrochemistry and Data Science working on first principles simulations of material properties, theoretical spectroscopy including quantum optimal control, and digital image processing.

Location:  Lecture Theatre 0.04 IMC

Networking Lunch: The Recharge Room, next to Lecture Theatre 004, from 12:30pm - 1pm.

Title: TBC

Abstract: TBC

Bio: Professor Mark Girolami is the Sir Kirby Laing Professor of Civil Engineering at the University of Cambridge, where he also holds the Royal Academy of Engineering Research Chair in Data-Centric Engineering. He is a Fellow of Christ’s College, Cambridge, and currently serves as Chief Scientist of The Alan Turing Institute, the UK’s national institute for data science and artificial intelligence.

A computational statistician by background, Professor Girolami’s research focuses on computational statistics, Bayesian methodology, probabilistic numerics, and their applications across engineering and the natural sciences. His projects span areas such as data-centric engineering, computational statistical inference for engineering and security, multimodal data analysis, digital twins, and urban systems modeling. He also plays a leading role in the Centre for Smart Infrastructure and Construction as Data Lead.

Before his appointment at Cambridge, Professor Girolami was Chair of Statistics at Imperial College London and one of the founding Executive Directors of The Alan Turing Institute, where he later became Strategic Programme Director and established the Lloyd’s Register Foundation Programme on Data-Centric Engineering. Earlier in his career, he spent a decade as a Chartered Engineer at IBM, before moving fully into academia.

His contributions have been widely recognized: he is a Fellow of the Royal Academy of Engineering and the Royal Society of Edinburgh, a past EPSRC Advanced and Established Career Fellow, and recipient of the Royal Society Wolfson Research Merit Award. In 2023, he was awarded the Guy Medal in Silver by the Royal Statistical Society. He has also delivered distinguished lectures including the IMS Medallion Lecture (2017), the Bernoulli Society Forum Lecture (2017), and the BCS & IET Turing Talk (2020).

Professor Girolami is also active in publishing and editorial leadership: he is Editor-in-Chief of Statistics and Computing and the founding Editor-in-Chief of Data-Centric Engineering (Cambridge University Press). Alongside his research, he is committed to teaching and mentoring, leading courses in Mathematical Methods and Computational Statistics and Machine Learning, and co-authoring the widely used textbook A First Course in Machine Learning.

Location: Lecture Theatre 0.04 IMC

Networking Lunch: The Recharge Room, next to Lecture Theatre 004, from 12:30pm - 1pm.

Title: Inference for State Space Models with Long Data by Variational Methods

Abstract: Statistical and machine learning methods allow practical inference for complex state space models (SSMs). However, standard approaches require sampling hidden states for the entire time range of interest in each training iteration, which can be impractically costly for larger datasets. Methods which use a shorter minibatch of data in each iteration are cheaper, but can introduce severe bias in a time series context. We avoid this by adapting the buffering approach of Aicher et al. (2019, 2025) to a variational inference context. We provide theoretical results supporting our approach, and empirical studies showing that the method achieves accurate results and orders of magnitude speed-ups.

Bio: Dennis Prangle is an associate professor in statistics at the University of Bristol.

His current research is on the interface between Bayesian statistics and machine learning. He is particularly interested in developing approximate inference methods such as simulation based inference approaches, variational inference and composite likelihood. One application is to likelihood-free inference, where simulation of data is possible but the likelihood function is unavailable. Another is to stochastic differential equations and he has worked on applications to population genetics, physics, ecology and epidemiology. He is also interested in experimental design and how to quickly derive effective high dimensional designs.

Location:  Lecture Theatre 0.04 IMC

Networking Lunch: The Recharge Room, next to Lecture Theatre 004, from 12:30pm - 1pm.

Title: TBC

Abstract: TBC

Bio: Prof. Marcin Kamiński has been a full professor in the Department of Structural Mechanics, Łódź University of Technology since 2015 and now He is the Head of the Discipline Civil Engineering, Geodesy & Transportation in the Faculty of Civil Engineering, Architecture & Environmental Engineering since 2019. He worked before in Division of Mechanics of Materials (1994-2009), was the Head of Department of Steel Structures (2010-2013); then He joined Department of Structural Mechanics as the Head of Division of Structural Reliability.

His main research includes the Stochastic Finite Element Method (and some other discrete stochastic numerical methods), numerical modeling of random composites including homogenization theory, reliability assessment and optimization of civil engineering structures, stochastic ageing processes, and also recently - an application of probabilistic entropy in engineering computations (in the framework of the research grant OPUS no. 2021/41/B/ST8/02432 sponsored by the National Science Center in Cracow, Poland).

He spent a postdoctoral study at Rice University in Houston, TX, USA, was a visiting professor at Leibniz-Institute of Polymer Research in Dresden, Germany, and also at Politecnico di Milano, Italy. He authored more than 200 research papers in various journals including International Journal for Numerical Methods in Engineering, Composite Structures, International Journal of Solids & Structures, and also Computers & Structures. He published two monographs: “Computational Mechanics of Composite Materials”, Springer, 2005, and also “The Stochastic Perturbation Method for Computational Mechanics”, Wiley, 2013. He serves as the associate editor in Mechanics Research Communications (by Elsevier, since 2013) and the member of editorial board in International Journal for Numerical Methods in Engineering (Wiley), Acta Mechanica (Springer), Computers & Structures (Elsevier), Journals of Composites Science and SCI (MDPI).

He was the recipient of the fellowships from Foundation for Polish Science (1996 & 1999), J. Argyris Award from ECCOMAS and Elsevier (2001), J.T. Oden Scholarship at The University of Texas (Austin, TX, USA, 2004), and also some Polish national or academic awards including Bronze Cross of Merit (1998), Silver Cross for the Longstanding Service (2017), Medal of Commission of Education (2022), and also very recently The Research Prize of Ministry of Education & Science in 2023. He has been recognized recently (since 2021) as the World’s Top 2% Scientists by Stanford University, CA, USA.

Prof. M. Kamiński teaches mainly the courses relevant to Structural Reliability, Optimization & Parametric Design and Finite Element Method, but also various courses related to Steel Structures, Strength of Materials and Theoretical Mechanics. He coauthored (with prof. B. Rogowski) the academic textbook “Technical Mechanics” having two editions – in 2009 & 2013. He taught in His career many academic courses for Civil Engineering, Environmental Engineering, Architecture & Architecture Engineering. He served as the promoter for 8 Ph.D. theses as well as for more than 40 M.Sc. and B.Sc. students.