Seminar Series

The Scientific Computing seminars usually take place on Wednesday at 16:00hrs in the Visualisation Lab (VisLab) (Room No.: MCS1022) located at the first floor of the MCS (Mathematical Sciences and Computer Science) building.

To sign up for the seminar mailing list (to receive updates or information on occasional virtual talks), please email [email protected].

If you are interested to share your research with us and would like to know our research activities please contact our SciCOMP Seminar Coordinator: Shounak Chakraborty [[email protected]]

We will be more than happy to welcome you here at SciComp, Durham.

Upcoming seminars

• Seminars for 2025/26 session will be started in October 2025.

Past seminars

Academic year 2024/2025

• Wednesday, 21 May 2025 (16:00 – 17:00hrs, UK Time), Dr. Timo Betcke, (University College London, United Kingdom) Location: MCS 1022 (VisLab)
  • Title: From 0 to Archer2 in 65k lines of Rust
  • Abstract: Rust is an exciting programming language. Within 10 years of its 1.0 release it has found its ways into Android, the Windows Kernel, cloud computing, and many other domains. But is it ready for the HPC Challenge? Three years ago we started off with 0 lines of code and the desire to develop a scalable Galerkin Boundary Element Code for Archer2. The caveat, we wanted to make it work in Rust. In this talk we give a breakdown of this journey, discuss things that worked really well in Rust, and other things that did not work so well. While the talk itself is not a Rust tutorial we want to give a realistic expectation of what expects those who want to do Rust at scale.
• Wednesday, 7 May 2025 (16:00 – 17:00hrs, UK Time), Dr. Julian Hall, (The University of Edinburgh, United Kingdom) Location: MCS 1022 (VisLab) [A joint event with NESTiD]
  • Title: Linear programming solvers old and new for HiGHS
  • Abstract: Linear programming is the fundamental problem in constrained optimization, so efficient solvers are very valuable. Until recently, there were two fundamental techniques: the simplex algorithm and interior point methods. However, first order methods are now offering a genuine third approach, in particular because they can exploit the rapid growth in GPU technologies. This talk will discuss the state-of-the-art in all three solution methods in the context of HiGHS, the world’s best open-source linear optimization software.
• Friday, 2 May 2025 (16:00 – 17:00hrs, UK Time), Dr. Ricardo Vinuesa, (KTH Engineering Mechanics, Sweden) Location: Online
  • Title: Improving turbulence control through explainable deep learning
  • Abstract: In this work we first use explainable deep learning based on Shapley explanations to identify the most important regions for predicting the future states of a turbulent channel flow. The explainability framework (based on gradient SHAP) is applied to each grid point in the domain, and through percolation analysis we identify coherent flow regions of high importance. These regions have around 70% overlap with the intense Reynolds-stress (Q) events in two-dimensional vertical planes. Interestingly, these importance-based structures have high overlap with classical turbulence structures (Q events, streaks and vortex clusters) in different wall-normal locations, suggesting that this new framework provides a more comprehensive way to study turbulence. We also discuss the application of deep reinforcement learning (DRL) to discover active-flow-control strategies for turbulent flows, including turbulent channels, three-dimensional cylinders and turbulent separation bubbles. In all the cases, the discovered DRL-based strategies significantly outperform classical flow-control approaches. We conclude that DRL has tremendous potential for drag reduction in a wide range of complex turbulent-flow configurations.
  • Speaker’s Bio: Dr. Ricardo Vinuesa is an Associate Professor at the Department of Engineering Mechanics, KTH Royal Institute of Technology in Stockholm. He is also Lead Faculty at the KTH Climate Action Centre. He studied Mechanical Engineering at the Polytechnic University of Valencia (Spain), and he received his PhD in Mechanical and Aerospace Engineering from the Illinois Institute of Technology in Chicago. His research combines numerical simulations and data-driven methods to understand, control and predict complex wall-bounded turbulent flows, such as the boundary layers developing around wings and urban environments. Dr. Vinuesa has received, among others, an ERC Consolidator Grant, the TSFP Kasagi Award, the MST Emerging Leaders Award, the Goran Gustafsson Award for Young Researchers, the IIT Outstanding Young Alumnus Award, the SARES Young Researcher Award and he leads several large Horizon Europe projects. He is also a member of the Young Academy of Science of Spain.
• Wednesday, 30 April 2025 (16:00 – 17:00hrs, UK Time), Dr. Prakash Murali, (University of Cambridge, United Kingdom) Location: MCS 1022 (VisLab)
  • Title:Using resource estimation to explore the scalability of quantum computer architectures
  • Abstract: There is a huge gap between the resource requirements of useful quantum computing (QC) applications and the hardware that is buildable now. My research seeks to close the applications-to-devices resource gap in QC by developing quantum computer architecture and compilation techniques. In this talk, I will present a resource estimation framework for understanding the qubit and runtime requirements of practically useful quantum applications. This framework provides a full-stack approach to model the quantum stack and offers lessons for choosing the appropriate type of qubits and error correction that can enable scalable computations. I will also present recent work from my group on optimizing the layers of the stack to reduce application resource requirements.
  • Speaker’s Bio: Dr. Prakash Murali is an associate professor in the Department of Computer Science, Cambridge University (UK). His research interests include quantum architecture, resource estimation and compilation. Prior to Cambridge, he was a Senior Quantum Systems Architect as part of Microsoft’s quantum computing program where he designed the Azure Quantum Resource Estimator to understand the resource needs of practical-scale quantum applications. Prakash graduated with a Computer Science Ph.D. from Princeton University. His work has been recognized by several awards, including the ACM SIGARCH/IEEE CS TCCA Outstanding Dissertation Award in 2022. 
• Monday, 28 April 2025 (16:00 – 17:00hrs, UK Time), Mr. Davide Villa, (Xinnor) Location: Online
  • Title: How to assure data resiliency at max performance for new AI and ML workloads
  • Abstract: AI and ML workloads require more performing storage solution than traditional HPC. Indeed, these workloads involve a combination of sequential and random access both in read and write operations. For this reason, all flash clusters are becoming more popular. Unfortunately, these new storage solutions are normally very expensive as they are based on proprietary hardware and software, with consequent strong vendor lock-in. In this talk, Xinnor will present its innovative RAID engine for NVMe drives, that can be used as a foundation to maximize resiliency and performance of traditional parallel file systems using standard hardware. We’ll explain how various prestigious Universities around the global used our technology to build highly resilient, performing and cost-effective storage clusters to address their AI and ML project.
• Wednesday, 19 March 2025 (16:00 – 17:00hrs, UK Time), Dr. Raymond S., (University of Saskatchewan, Canada) Location: MCS 1022 (VisLab) [Virtual]
  • Title: Stability and efficiency enhancements of operator-splitting methods
  • Abstract: Operator-splitting methods are widely used for the time integration of differential equations, especially those that arise from multi-scale or multi-physics models, because a monolithic approach may be inefficient or even infeasible. The most common operator-splitting methods are the first-order Lie–Trotter (or Godunov) and the second-order Strang (Strang–Marchuk) splitting methods. High-order splitting methods with real coefficients require backward-in-time integration in each operator and hence may be impacted by instability. However, besides the methods themselves, there are many other ancillary aspects to an overall operator-splitting method that are important in practice but often overlooked. For example, the order in which operators are integrated and the choice of sub-integration methods can significantly affect the performance of an operator-splitting method. In this paper, we design a new four-stage, third-order, 2-split operator-splitting method with seven sub-integrations and an optimized linear stability region. We then propose two general strategies to further improve its stability and efficiency for a specific problem, namely, to choose the ordering of operators to maximize linear stability and to choose low-order explicit sub-integrators for unstable sub-integrations.  We demonstrate about a 40% improvement in the performance from the combined use of these strategies relative to standard implementations on a benchmark problem from cardiac electrophysiology.
• Friday, 14 March 2025 (16:00 – 17:00hrs, UK Time), Dr. Jose Cano Reyes, (University of Glasgow, United Kingdom) Location: MCS 1022 (VisLab)
  • Title: Accelerating AI at the Edge: The Power of Efficient Hardware-Software Co-Design
  • Abstract: Deep Neural Networks (DNNs) are increasingly a key component within Artificial Intelligence (AI) applications for a number of domains, including computer vision, natural language processing, and scientific computing. At the same time, executing DNN models on edge devices may allow secure computation and lower energy consumption and cost, but to become practical performance must improve dramatically. This is due to the significant demands introduced by emerging DNN models in terms of both memory and compute and the reduce availability of them in constrained edge devices. In this talk I will introduce the Glasgow Intelligent Computing Laboratory (gicLAB) and give an overview of our current and future research, with an emphasis on Hardware/Software co-design approaches to efficiently deploy and run AI/ML applications on constrained edge devices. 
  • Speaker’s Bio: José Cano is an Associate Professor in the School of Computing Science at the University of Glasgow, where he leads the Glasgow Intelligent Computing Laboratory (gicLAB) within the Systems Research Section (GLASS), and is deputy Head of GLASS. His research interests are in the broad areas of Computer Architecture, Computer Systems, Compilers, Machine Learning, and Security. His current research is focused on Edge AI, and more specifically on ‘Hardware-software co-design approaches for efficient deployment and execution of AI applications on mobile/embedded edge devices’. José is currently Principal Investigator at the University of Glasgow on the EU’s Horizon Europe project dAIEDGE, and Co-Investigator on the UKRI “Digital Security by Design” projects AppControl and Morello-HAT. He was Principal Investigator on the UK’s PETRAS project MAISE. He has obtained >£550K as a project PI, >£1.3M as a project Co-I, and >£81K from six personal grants. José received his Ph.D. in Computer Science from Universitat Politècnica de València (Spain) in January 2012. After that he was a Postdoctoral Researcher in the Department of Computer Architecture at Universitat Politècnica de Catalunya (Spain) until December 2013. Then he joined the Institute for Computing Systems Architecture in the School of Informatics at The University of Edinburgh (UK) where he was a Research Associate between January 2014 and August 2018. He is a senior member of the IEEE and ACM research societies and a member of the HiPEAC, dAIEDGE and PETRAS networks of excellence.
• Wednesday, 12 March 2025 (16:00 – 17:00hrs, UK Time), Dr. Marta Betcke, (University College London, United Kingdom) Location: MCS 1022 (VisLab) [Virtual]
  • Title: Machine learning in solution of inverse problems: subjective perspective
  • Abstract: Following the 2012 breakthrough in deep learning for classification and visions problems, the last decade has seen tremendous raise of interest in machine learning in a wider mathematical research community from foundational research through field specific analysis to applications. As data is at the core of any inverse problem, it was a natural direction for the field to investigate how machine learning could aid various aspects of inversion yielding numerous approaches from somewhat ad-hoc but very effective like learned unrolled methods to provably convergent learned regularisers with everything in between. In this talk I will review some on these developments through a lens of the research of our group. 
• Wednesday, 26 February 2025 (16:00 – 17:00hrs, UK Time), Dr. Matthieu Schaller, (Leiden University, The Netherlands)Location: MCS 1022 (VisLab)
  • Title: Running large cosmological simulations & developing large HPC applications for users
  • Abstract: In the modern era, numerical simulations are the key for astrophysics and cosmology theorists. The running and analysis of complex multi-physics simulations is now common practice and modern codes are able to exploit the largest HPC systems in existence. One of the fairly unique challenges intrinsic to the nature of cosmological time-steps is the localised time-stepping algorithms which make strong- and weak-scaling of codes a challenging task. In this talk, I will discuss some of the building pieces and design choices behind the SWIFT simulation code, the package used to run the largest cosmological simulation to date. I will also discuss the additional complexities created by the complex needs of application scientists and the design compromises this leads to. The best algorithm is not always the best choice for the end user’s needs. I will conclude with some thoughts on the step forwards to future-proof the code for new architectures, with an emphasis on the work undertaken by the Durham team.
  • Speaker’s Bio: Matthieu Schaller is an assistant professor at Leiden University working on the development, design, running, and analysis of cosmological galaxy formation simulation. He is the lead-developer of the SWIFT simulation code used for multiple large campaigns of simulations, including the largest simulation ever run. His expertise spans cosmology and high-performance computing with contributions to numerical methods as well as algorithmic and software engineering for large user communities. He has interests in large-scale magnetic fields, magnetogenesis scenarios, baryon effects on late-time cosmology probes, astrophyiscal probes of the nature of dark matter, galaxy formation and evolution, numerical methods, fluid dynamics, high-performance computing & cheese.
• Thursday, 20 February 2025 (16:00 – 17:00hrs, UK Time), Dr. Jefferey S. Young (Georgia Institute of Technology, USA) Location: MCS 1022 (VisLab)
  • Title: Novel Architecture Infrastructure for Addressing Future HPC Data Challenges
  • Abstract: As we move through the Exascale era and look to integrate more data-intensive AI capabilities into our existing HPC workflows, one key question remains: How will we address the challenges of data movement that will drive future performance gains. This talk will cover our collaborative work at Georgia Tech on building a novel “post-Moore” testbed, the CRNCH Rogues Gallery, as well as three case studies focused on the following areas: 1) Challenges in developing and deploying near-memory systems to address the demands of sparse, graph-based applications; 2) Opportunities for better evaluating future memory system performance using the open-source Spatter benchmark; and 3) The emergence of SmartNICs and in-network computing as a future avenue for HPC domain-specific data accelerators.
  • Speaker’s Bio: Jeffrey Young is a principal research scientist with Georgia Tech’s Partnership for Advanced Computing Environments (PACE). With a background in computer architecture, his main research interests have focused on the intersection of high-performance computing and novel accelerators including GPUs, FPGAs, SmartNICs, and next-generation processors. He is the director of a novel architecture testbed, the CRNCH Rogues Gallery, that aims to simplify and democratize access to novel post-Moore accelerators in the neuromorphic, reversible, and smart networking spaces. Dr. Young also co-directs Georgia Tech’s Center for Scientific Software Engineering and is the director of GT’s Open Source Program Office. He received his PhD in computer engineering in 2013 from Georgia Tech’s ECE department.
• Wednesday, 19 February 2025 (16:00 – 17:00hrs, UK Time), Prof. Alex Yakovlev, (Newcastle University, UK) Location:  MC Maple 1 & MC Maple 2 (Mountjoy)
  • Title: Tsetlin Machines: stepping towards energy-efficient, explainable and dependable AI
  • Abstract: Artificial Intelligence (AI) and Machine Learning (ML) enter our lives in many forms, from high-end data processing and mining for applications such as medical diagnosis and cyber-commerce to low-end intelligent interfaces (mobile and internet of things (IoT) devices) for voice and image recognition applications such as industrial and household sensing and healthcare monitoring. Lately, ML has been gradually albeit cautiously (!) entering safety-critical applications. The key challenges on this path are the issues of, firstly, high cost of conventional ML methods, such as deep learning (e.g. DNNs), in terms of energy and computational resources, and secondly, the lack of interpretability of the models. Tsetlin Machine (TM) is a recent logic and automaton-based model for reinforcement learning. It has demonstrated competitive accuracy on many popular benchmarks while providing a natural interpretability as well as energy-efficiency, enabling this model for both inference and training at the edge. The talk will provide an overview of TM architecture and its parameter tuning. The gains in energy-efficiency and interpretability, and hence trustworthiness, against DNNs will be illustrated through a number of case studies.
  • Speaker’s Bio: Alex Yakovlev, PhD (1982), DSc (2006). Since 1991 he is with Newcastle University, UK, where he is a Professor of Computer Systems Design, founded and leads the Microsystems Research Group, and co-founded the Asynchronous Systems Laboratory. He was awarded an EPSRC Dream Fellowship in 2011–2013. He has published 8 edited and co-authored monographs and more than 500 papers in IEEE/ACM journals and conferences, in the areas of concurrent and asynchronous circuits and systems, Petri nets, electronic design automation, low power circuits and systems, AI and machine learning hardware based on Tsetlin automata and electromagnetic computing, with several best paper awards and nominations. He co-invented Signal Transition Graphs (STGs) and co-led developments of tools for them (Petrify, Workcraft) throughout the last 30 years. He has supervised over 70 PhD students. He is a Fellow of Royal Academy of Engineering and Fellow of IEEE. He is a co-founder of a recently created spin-out company Literal Labs (formerly Mignon Technologies), commercialising solutions for ML at the edge.
• Tuesday, 18 February 2025 (16:00 – 17:00hrs, UK Time), Prof. Richard W. Vuduc, (Georgia Institute of Technology, USA) Location: MCS 1022 (VisLab)
  • Title: Are AI machines good for HPC?
  • Abstract: Supercomputer architectures are being dominated by a single workload: AI training. Is that good for high-performance computing (HPC) more broadly? This talk speculates on the relative merits—and pitfalls—of “AI machines” for HPC workloads based on a we developed called Calculon, which aims to facilitate algorithm-architecture codesign for large language models via a high-level analytical performance model.
  • Speaker’s Bio: Richard (Rich) Vuduc is a professor at Georgia Tech in the School of Computational Science and Engineering. His research lab, the HPC Garage, is interested in performance “by any means necessary,” whether by more innovative algorithms, better analysis, more effective programming techniques, or novel hardware.
• Monday, 17 February 2025 (16:00 – 17:00hrs, UK Time), Prof. Edmond T. Chow, (Georgia Institute of Technology, USA) Location: MCS 1022 (VisLab)
  • Title: Kernel Matrices: From Physics to Machine Learning
  • Abstract: Kernel matrices, defined by a set of points and a pairwise interaction function, have garnered significant attention recently due to rising interest in Gaussian process regression and other kernel methods in machine learning. However, kernel matrices have a long history, particularly in computational physics and integral equation problems, often under different names. In machine learning, kernel methods are often perceived as limited by the computational cost of processing the data, primarily due to the need to solve systems of equations involving the kernel matrix. Recently, the intersection of algorithms from physical applications and statistical ideas has led to innovative methods for kernel matrix problems. In this presentation, we will explore the hierarchical approximation of kernel matrices, enabling storage and operations to be performed in linear time relative to the number of points. We will also present a preconditioner designed for the iterative solution of kernel matrix systems, specifically targeting Gaussian process hyperparameter estimation.
  • Speaker’s Bio: Edmond Chow is Professor and Associate Chair in the School of Computational Science and Engineering at Georgia Institute of Technology. His research is in developing numerical methods specialized for high-performance computers and applying these methods to enable the solution of large-scale physical simulation problems in science and engineering. Dr. Chow previously held positions at D. E. Shaw Research and Lawrence Livermore National Laboratory. He was Chair of the 2022 ACM Gordon Bell Prize committee, and was Co-Chair of the 2022 SIAM Annual Meeting. He is currently Vice-Chair of the SIAM Activity Group on Computational Science and Engineering. Dr. Chow is a Fellow of SIAM.
• Wednesday, 12 February 2025 (16:00 – 17:00hrs, UK Time), Dr. Eleni Vlachopoulou, (AMD, Manchester, UK) Location: MCS 1022 (VisLab)
  • Title: Optimizing k-Nearest Neighbours Algorithm
  • Abstract: The AOCL-Data Analytics Library (AOCL-DA) provides optimized building blocks for data analysis and classical machine learning, including a wide range of algorithms such as linear models, k-means clustering, principal component analysis, and nonlinear least squares fitting. Additionally, AOCL-DA offers an extension to scikit-learn, allowing scikit-learn users to seamlessly take advantage of the library. In this seminar, we will focus on the k-nearest neighbours algorithm, a commonly-used machine learning method that is particularly effective when there are a small number of predictors. We will delve into the theoretical foundations of the Brute Force algorithm, discuss practical considerations for the software design, and present the optimizations we implemented in AOCL-DA to improve the performance of k-nearest neighbours classifiers when working with large datasets.
• Wednesday, 5 February 2025 (16:00 – 17:00hrs, UK Time), Dr. Andrew Valentine, (Durham University, UK) Location: MCS 2068
  • Title: Solving inference problems without assumptions
  • Abstract: In geophysics—and in all science—the questions we ask determine the answers we get. Every analysis we perform is underpinned by a host of assumptions and arbitrary choices: Which data should we use? What will we regard as ‘unknowns’, and what will we pretend to know exactly? Is Earth a perfect sphere, an oblate spheroid, or maybe just an infinite half-space? All of these decisions have an impact on the detailed results that we obtain, and perhaps even on the conclusions that we will eventually reach. Yet making such choices is unavoidable — or is it? In this talk, I will show two recent developments that are each designed to remove assumptions from geophysical imaging and inference. The first is the concept of ‘overcomplete tomography’, introduced by Turunçtur et al. (2023): by over-parameterising and exploiting the idea of ‘sparsity’, we are able to avoid imposing pre-determined characteristics and length-scales upon tomographic images. Second, ‘trans-conceptual sampling’ (Sambridge et al, in review) extends this idea into the domain of Bayesian inference, and enables quantitative assessment of competing theoretical frameworks against observational data.
• Wednesday, 11 December 2024 (16:00 – 17:00hrs, UK Time), Dr. Hatem Ltaif, (KAUST, Saudi Arabia), Location: MCS 1022 (VisLab) [A SIAM Supercomputing Spotlights Webinar]
  • Title: Solving Big Problems with Little Numbers
  • Abstract: The future of simulations lies in leveraging hardware features designed for the AI market, particularly in low-precision computations. Modern NVIDIA GPUs exemplify this trend, offering significant performance gains through low-precision computations, resulting in reduced elapsed time, smaller memory footprints, and energy savings. We harness these capabilities to develop fast mixed-precision linear algebra algorithms. Our adaptive precision conversion strategy dynamically adjusts computation accuracy, maintaining high precision only where necessary within the matrix operator, while still meeting application-worthy precision requirements. This talk will illustrate how these algorithms revolutionize computational efficiency for geospatial statisticians, bioinformaticians, and geophysicists, having significant implications for environmental computational statistics, genome-wide association studies in computational biology, and seismic imaging for CO2 sequestration.
• Wednesday, 4 December 2024 (16:00 – 17:00hrs, UK Time), Dr. Corentin Houpert, (School of Computing and Mathetical Sciences, The University of Leicester, UK), Location: MCS 1022 (VisLab)
  • Title: Physics-Informed Autoencoder for Enhancing Data Quality to Improve the Forecasting Reliability of Carbon Dioxide Emissions from Agricultural Fields
  • Abstract: Missing values in measurements for carbon dioxide emissions on drained peatlands remains an open challenge for training forecasting techniques to achieve net zero. At the field scale, existing methods struggle to model CO_2 emissions to fill gaps, especially in nighttime measurements. We propose robust Physics-Informed Autoencoders (PIAEs), which combine the generative capabilities of Autoencoders with the reliability of physical models of Net Ecosystem Exchange (NEE) that quantify CO_2 exchanges between the atmosphere and major carbon pools. Our method integrates equations describing the physical processes and associated uncertainties to fill gaps in NEE measurements from eddy covariance (EC) flux towers. In the PIAE, various sensor measurements are encoded into the latent space, and a set of decoders is then used to approximate the ecosystem parameters and the optimal NEE forecast, directed by dynamics described by a stochastic differential equation. These decoders utilize nighttime and daytime NEE models that describe carbon transfer as a Wiener process. Finally, we use a two-phased training routine with two loss functions describing each phase: Mean Squared Error (MSE) and Maximum Mean Discrepancy (MMD) between the measurements and the reconstructed samples. PIAE outperforms the current state-of-the-art Random Forest Robust on the prediction of nighttime NEE measurements on various distribution-based and data-fitting metrics. We present significant improvement in capturing temporal trends in the NEE at daily, weekly, monthly and quarterly scales.
• Wednesday, 27 November 2024 (16:00 – 17:00hrs, UK Time), Dr. Rui Carvalho, (School of Engineering, Durham University) Location: MCS 1022 (VisLab)
  • Title: Automatically Extracting Partial Differential Equations from Data
  • Abstract: Identifying partial differential equations (PDEs) from data is crucial for understanding the governing mechanisms of natural phenomena, yet it remains a challenging task. We present an extension to the ARGOS framework, ARGOS-RAL, which leverages sparse regression with the recurrent adaptive lasso to identify PDEs from limited prior knowledge automatically. Our method automates calculating partial derivatives, constructing a candidate library, and estimating a sparse model. We rigorously evaluate the performance of ARGOS-RAL in identifying canonical PDEs under various noise levels and sample sizes, demonstrating its robustness in handling noisy and non-uniformly distributed data. We also test the algorithm’s performance on datasets consisting solely of random noise to simulate scenarios with severely compromised data quality. Our results show that ARGOS-RAL effectively and reliably identifies the underlying PDEs from data, outperforming the sequential threshold ridge regression method in most cases. We highlight the potential of combining statistical methods, machine learning, and dynamical systems theory to automatically discover governing equations from collected data, streamlining the scientific modelling process.
• Wednesday, 20 November 2024 (16:00 – 17:00hrs, UK Time), Prof. Rishad Shafik, (School of Engineering, Newcastle University) Location: MCS 1022 (VisLab)
  • Title: Empowering Logic Driven AI Systems at the Edge using Tsetlin Machine
  • Abstract: Enabling artificial intelligence (AI) at the edge has the potential to revolutionise a new generation of autonomous applications. Achieving this vision requires energy efficiency to be a primary focus in both hardware and software system design. In this talk, I will introduce a novel AI ecosystem underpinned on the Tsetlin Machine (TM), which represents a shift from traditional arithmetic-based AI to logic-based AI. I will outline the foundational principles of this approach and showcase key advancements, including several TM microchip designs, innovative data preprocessing techniques, self-timed and online learning microarchitectures, and real-world applications. Finally, I will discuss the challenges encountered to date and explore the opportunities for future research in this promising area.
  • Speaker’s Bio: Professor Rishad Shafik (RS) is a Personal Chair in Microelectronic Systems Design and EEE Research Director of EEE at Newcastle University. He is an international leader of hardware/software co-design applied in machine learning systems. He has published in excess of 200 research articles in major peer-reviewed IEEE/ACM journals and conferences, with 4 of them winning the best paper awards and 4 others nominated for best paper awards. His research contributed to circa £29m research grants as PI/CoI funded by EPSRC, Research Council of Norway (RCN) and Industries. Underpinned on two recent patents and £500k accelerator grants from EPSRC and Research England, he has recently founded Literal Labs AI (a Newcastle University spinout specialising in ML co-processor architectures and embedded solutions).  
• Wednesday, 13 November 2024 (16:00 – 17:00hrs, UK Time), Dr. Ben Wooding, (School of Computing, Newcastle University) Location: MCS 1022 (VisLab)
  • Title: IMPaCT: A software tool for controller synthesis of stochastic systems using interval Markov decision processes
  • Abstract: In this presentation I will discuss an open-source software tool, called IMPaCT, for the parallelized verification and controller synthesis of large-scale stochastic systems using interval Markov chains (IMCs) and interval Markov decision processes (IMDPs). Controllers designed using IMPaCT enable strong guarantees to be provided on the behavior of the stochastic system under analysis over finite and infinite-horizon properties, including safety, reachability, and reach-avoid. As part of my talk I will introduce stochastic control systems and the technique of finite abstraction used within the formal control community to provide these strong guarantees. I will also highlight why the more common Markov decision processes (MDPs) are not viable for this controller design.
• Friday, 8 November 2024 (14:00 – 15:00hrs, UK Time), Dr. Juliette Dubois, (RWTH Aachen University, Germany) Location: MCS 2068
  • Title: Two topics on hyperbolic equations for geophysical modelling
  • Abstract: In my talk, I will present two projects focused on hyperbolic equations for geophysical modelling. The first project concerns the modelling of hydro-acoustic waves for tsunami early-warning systems. I will present the derivation of a wave-like linear model, its connection to other well-known models for acoustic waves and water waves, and some simulations. This work is a collaboration with Jacques Sainte-Marie, Sébastien Imperiale and Anne Mangeney. The second project concerns uncertainty quantification for nonlinear hyperbolic systems. The objective is to assess the dependency of the solution to a hyperbolic equation (for example the shallow water equation) to uncertainty in the initial condition. We use the stochastic finite volume method to compute the solutions depending on a large number of random parameters. In order to keep a reasonable computational cost, we combine the finite volume method with a low-rank approximation of the unknowns. This work is a collaboration with Michael Herty and Siegfried Müller. 
• Wednesday, 6 November 2024 (16:00 – 17:00hrs, UK Time), Dr. Laura Scarabosio, (Radboud University, The Netherlands) Location: MCS 1022 (VisLab)
  • Title: Forward and inverse shape uncertainty quantification with partial differential equations
  • Abstract: We consider the task of quantifying the effect of geometric uncertainties on the behavior of a system whose physical state is described by a partial differential equation. In particular, we focus on uncertainty in the shape of the physical domain or of an internal interface. We first address how such uncertainties can be modeled, and how to efficiently compute different realizations of the solution to the PDE. Then, we will address both the forward propagation of uncertainty and the inverse problem in a Bayesian setting. For both cases, we will discuss computational methods for efficient shape uncertainty quantification and their theoretical guarantees.
• Wednesday, 23 October 2024  (16:00 – 17:00hrs, UK Time), Dr. Hossein Amini Kafiabad, (Department of Mathematical Sciences, Durham University), Location: MCS 1022 (VisLab)
  • Title: The Curse and Blessing of Multiple Timescales in Simulating Geophysical Flows
  • Abstract: In this talk, I will cover a few interconnected projects with the underlying theme of the advantages and disadvantages of having both fast and slow motions in fluid dynamics. In the first part of my talk, I will start by highlighting the benefits of computing Lagrangian means for modeling, flow decomposition, and post-processing simulation data. Despite these benefits, computing Lagrangian means is challenging for large simulations. Typical implementations require tracking a large number of particles to construct Lagrangian time series, which are then averaged. This approach has drawbacks, including large memory demands, particle clustering, and complications with parallelisation. I will introduce a novel approach in which the Lagrangian means of flow variables are computed without tracking particles in time. This newly proposed algorithm can compute the Lagrangian means on-the-fly with the simulation to minimise the memory footprint, and it is more suitable for parallel implementations. In the second part of my talk, I will present our new idea for improving the numerical phase averaging method, which allows for larger time steps when solving systems with oscillatory stiffness. The phase averaging method is a strong candidate for use as a coarse propagator in the Parareal algorithm, which parallelises initial value problems in time. However, phase averaging itself is a costly operation in large simulations. I will explain how a specific averaging operator, defined with a relaxation time, can transform the averaging process into the solution of a differential equation that can be solved in parallel with the phase-averaged equations. This approach can substantially reduce the cost of phase averaging.
• Wednesday, October 2, 2024 (16:00 – 17:00hrs, UK Time), Dr. Alastair Basden (Department of Physics, Durham University), Location: MCS 1022 (VisLab)
  • Title:The Durham HPC Hardware Laboratory
  • Abstract: The Durham HPC Hardware Lab is hosted by the DiRAC COSMA HPC facility and provides UK researchers with access to cutting edge technologies and facilities, to allow testing of codes, software migration to new hardware, and study of new paradigms.  This lab has grown in scope over the past few years, funded by ExCALIBUR H&ES, DiRAC, Durham and various UKRI grants. Of particular interest to many users is access to new GPU systems, novel networking topologies, composable infrastructure, storage systems and access to BlueField DPUs. In addition to compute hardware, the Hardware Lab includes data centre-scale technologies, such as solar power generation, immersion cooling and waste heat storage. This talk presents the Hardware Lab, including information about how it can be accessed and used. 

Academic year 2023/2024

• Friday, 8 March 2024, 13:00, David Keitel, University of the Balearic Island, MCS 2051
  
  The event takes place in person in the VisLab of the MCS building, to attend online use the following zoom link: https://durhamuniversity.zoom.us/j/98751452277?pwd=ckpzVlc4TCtiQjZJWERwc2R1UGd6dz09
  Meeting ID: 987 5145 2277, Passcode: 002464
  
  Title: Gravitational-wave astronomy: what we’ve found so far and what we’re still looking for
  
  Abstract: The gravitational-wave window onto the Universe has been opened with the first detection of a black-hole merger in 2015. Since then, the LIGO-Virgo-KAGRA Collaboration has published 90 probable detections from three complete observing runs of the advanced-generation laser-interferometric detectors. These have enabled many new insights into the astrophysics of compact objects and the evolutionary history of massive stars, and are a completely novel probe for cosmology and fundamental physics. With the currently ongoing fourth observing run, future detector upgrades and completely new observatories, we will be able to reach much deeper into our Universe’s population of merging compact objects. But we are also still hunting for many more types of first detections, including gravitationally lensed gravitational waves as well as signals from spinning neutron stars or supernovae.
• Friday, 1 December 2023, 13:00, Filippo Spiga, NVIDIA – MCS2050 (joint with NESTiD)
  
  Title: The NVIDIA superchip (Grace-Grace and Grace-Hopper) platform: the ‘what’, the ‘how’, the ‘why’
  
  The purpose of this talk is to introduce the NVIDIA Grace CPU Superchip and NVIDIA Grace Hopper Superchip (CPU+GPU) platforms and how advancements in hardware coupled with NVIDIA’s vision on programming models for accelerated computing can have profound implications in developing next generation fast (time to solution) and efficient (energy to solution) HPC and AI codes.
  
  This event is an in-person meeting in the Computer Science department. However, there will be Zoom option offered through the NESTiD seminar.
  
• Thursday, 23 November 2023, 13:00, Philip Maybank, AMD
  
  MCMC for Bayesian Uncertainty Quantification from Time-Series Data
  
  In computational neuroscience, Neural Population Models (NPMs) are mechanistic models that describe brain physiology in a range of different states. Within computational neuroscience there is growing interest in the inverse problem of inferring NPM parameters from recordings such as the EEG (Electroencephalogram). Uncertainty quantification is essential in this application area in order to infer the mechanistic effect of interventions such as anaesthesia.
  
  This talk presents  software for Bayesian uncertainty quantification in the parameters of NPMs from approximately stationary data using Markov Chain Monte Carlo (MCMC). Modern MCMC methods require first order (and in some cases higher order) derivatives of the posterior density. The software presented offers two distinct methods of evaluating derivatives: finite differences and exact derivatives obtained through Algorithmic Differentiation (AD). For AD, two different implementations are used: the open source Stan Math Library and the commercially licenced  tool distributed by NAG (Numerical Algorithms Group). The use of derivative information in MCMC sampling is demonstrated through a simple example, the noise-driven harmonic oscillator. And different methods for computing derivatives are compared. The software is written in a modular object-oriented way such that it can be extended to derivative based MCMC for other scientific domains.
  
  The event takes place in person in the VisLab of the MCS building, to attend online use the following zoom link: https://durhamuniversity.zoom.us/j/94668043904?pwd=Y0czQjBKb2NTT2xtZkp5WlBxeTFOUT09

Academic year 2022/2023

Summer seminars

• Wednesday, 19 July 2023, at 16:00, Lukas Krenz, Technical University of Munich
  
  Title: The Power of Oomph: A loud story about earthquakes, tsunamis, sound, and HPC
  
  Abstract: This talk explains the implementation of elastic-acoustic coupling in the open-source software SeisSol. We discuss two applications with real-world scenarios: First, we introduce the Palu, Sulawesi, 2018 earthquake-tsunami event and present a fully-coupled model that captures the complete event from dynamic earthquake rupture, to wave propagation in the Earth and the ocean, to tsunami propagation. Tsunami propagation is included by using a linearized boundary condition. As a second scenario, we discuss earthquakes induced by an enhanced geothermal system in the Helsinki metropolitan area. We model the largest of these earthquakes and the audible sound excited by it. Finally, we discuss how applying local time stepping (LTS) leads to efficient simulations. We investigate a novel implementation of LTS using state machines. We present strong-scaling results for the fully-coupled Palu scenario on the Mahti and Frontera supercomputers.

Easter term

• Wednesday, 21 June 2023, at 16:00, Bora Uçar, CNRS and ENS Lyon, France
  
  Title: On the Birkhoff–von Neumann decomposition
  
  Abstract: The Birkhoff–von Neumann decomposition expresses a doubly stochastic matrix as a convex combination of permutation matrices. This talk will be an introduction to this decomposition. We will cover algorithms, combinatorial problems, and some open problems.
  
  This talk contains results from joint work with Michele Benzi (Scuola Normale Superiore, Pisa, Italy), Jérémy E. Cohen (CNRS, Lyon), Fanny Dufosse (Inria, France), Kamer Kaya (Sabanci Univ, Turkey), and Ioannis Panagiotas (LIP6, Sorbonne Univ., France).

• Wednesday, 7 June 2023, at 16:00, David Silvester, The University of Manchester
  
  Title: Fast solution of incompressible flow problems with two-level pressure approximation
  
  Abstract: Reliable and efficient iterative solvers for models of steady incompressible flow emerged in the early 1990s. Strategies based on block preconditioning of the underlying matrix operators using (algebraic or geometric) multigrid components have proved to be the key to realising mesh independent convergence (and optimal complexity) without the need for tuning parameters, particularly in the context of classical mixed finite element approximation. The focus of this contribution is on efficient solver strategies in cases where (an inf–sup) stable Taylor–Hood mixed approximation is augmented by a piecewise constant pressure in order to guarantee local conservation of mass. The augmentation leads to over-specification of the pressure solution requiring a redesign of the established solver technology.
  
  This enrichment process causes over-specification of the pressure, which complicates the design and implementation of efficient solvers for the resulting linear systems. We first describe the impact of this choice of pressure space on the matrices involved. Next, we show how to recover effective solvers for Stokes problems, using a preconditioner based on the singular pressure mass matrix, and for Oseen systems arising from linearising the Navier–Stokes equations, by using a two-stage pressure convection–diffusion strategy.
  
  This is joint work with Jennifer Pestana.

• Friday, 19 May 2023, at 14:00, Dan Stanzione, Texas Advanced Computing Center (TACC)
  
      Part of the Durham HPC Days – Spring 2023.
  
      Unusual time: Friday at 14:00
  
      Unusual venue: Scott Logic Lecture Theatre (MCS0001).
  
      Unusual online access: Zoom link for the HPC Days – Spring 2023
  
  Title: What’s going on in research computing and AI in the US and Texas
  
  Short bio: Dr. Dan Stanzione, Associate Vice President for Research at The University of Texas at Austin since 2018 and Executive Director of the Texas Advanced Computing Center (TACC) since 2014, is a nationally recognized leader in high performance computing. He serves on the National Artificial Intelligence Research Resource Task Force, formed by the National Science Foundation (NSF) and the White House Office of Science and Technology Policy (OSTP). He is the principal investigator (PI) for an NSF grant to deploy Frontera, the fastest supercomputer at any U.S. university. Stanzione is also the PI of TACC’s Stampede2 and Wrangler systems, supercomputers for high performance computing and for data-focused applications, respectively. For six years he was co-PI of CyVerse, a large-scale NSF life sciences cyberinfrastructure. Stanzione was also a co-PI for TACC’s Ranger and Lonestar supercomputers, large-scale NSF systems previously deployed at UT Austin. Stanzione received his bachelor’s degree in electrical engineering and his master’s degree and doctorate in computer engineering from Clemson University.

• Friday, 19 May 2023, at 13:00, David Keyes, King Abdullah University of Science and Technology
  
      Part of the Durham HPC Days – Spring 2023.
  
      Unusual time: Friday at 13:00
  
      Unusual venue: Scott Logic Lecture Theatre (MCS0001).
  
      Unusual online access: Zoom link for the HPC Days – Spring 2023
  
  Title: Efficient computation through tuned approximation
  
  Abstract: Numerical linear algebra software is being reinvented to provide opportunities to tune dynamically the accuracy of computation to the requirements of the application, resulting in savings of memory, time, and energy. Floating point computation in science and engineering has a history of “oversolving” relative to expectations for many models. So often are real datatypes defaulted to double precision that GPUs did not gain wide acceptance until they provided in hardware operations not required in their original domain of graphics. Indeed, the condition number of discretizations of the Laplacian reaches the reciprocal of unit roundoff for single precision with just a thousand uniformly spaced points per dimension. However, many operations considered at a blockwise level allow for lower precision and many blocks can be approximated with low rank near equivalents. This leads to smaller memory footprint, which implies higher residency on memory hierarchies, leading in turn to less time and energy spent on data copying, which may even dwarf the savings from fewer and cheaper flops. We provide examples from several application domains, including a review of a 2022 Gordon Bell finalist computation that benefits from both blockwise lower precisions and lower ranks.
  
  Short bio: David Keyes directs the Extreme Computing Research Center at the King Abdullah University of Science and Technology (KAUST), where he was a founding Dean in 2009 and currently serves in the Office of the President as Senior Associate. He is a professor in the programs of Applied Mathematics, Computer Science, and Mechanical Engineering. He is also an Adjunct Professor of Applied Mathematics and Applied Physics at Columbia University, where he formerly held the Fu Foundation Chair. He works at the interface between parallel computing and PDEs and statistics, with a focus on scalable algorithms that exploit data sparsity. Before joining KAUST, Keyes led multi-institutional scalable solver software projects in the SciDAC and ASCI programs of the US Department of Energy (DoE), ran university collaboration programs at US DoE and NASA institutes, and taught at Columbia, Old Dominion, and Yale Universities. He is a Fellow of SIAM, the AMS, and the AAAS. He has been awarded the Gordon Bell Prize from the ACM, the Sidney Fernbach Award from the IEEE Computer Society, and the SIAM Prize for Distinguished Service to the Profession. He earned a B.S.E. in Aerospace and Mechanical Sciences from Princeton in 1978 and a Ph.D. in Applied Mathematics from Harvard in 1984.

• Thursday, 18 May 2023, at 13:00, Emma Barnes, University of York
  
      Part of the Durham HPC Days – Spring 2023.
  
      Unusual time: Thursday at 13:00
  
      Unusual venue: Scott Logic Lecture Theatre (MCS0001).
  
      Unusual online access: Zoom link for the HPC Days – Spring 2023
  
  Title: Sustainable accessible research IT
  
  Short bio: Emma Barnes is Head of Research IT at the University of York. Emma has spent the last 8 years building the research IT offering at the University. Emma project managed the first major cluster offering at the University (Viking). The £2.5 million project offers researchers and academics free access to the technology, and has been a huge success with users from a range of disciplines and backgrounds. We are now working on its replacement with a bigger focus on sustainability. The research IT team has also recently established a Research Software Engineering group and her focus is now on building up the infrastructure team where we can promote career development, training and peer support. The team’s other focus is accessibility, either through educating users or embracing new technologies. Where we are now focusing on efforts to support non- traditional HPC users and where appropriate, implementing new technologies to enhance research and teaching. Emma received her MPhys in Physics with Astrophysics at the University of York, then completed her PhD in Astroparticle physics at the University of Edinburgh. Edinburgh was where Emma became a programming and Linux enthusiast, which continued throughout her Postdoctoral work in Boston University US in Particle physics. Emma later switched careers to a more computing focus and can now use her passion for research IT to benefit research throughout the university.

• Wednesday, 17 May 2023, at 16:00, François Mazen, Kitware
  
      Part of the Durham HPC Days – Spring 2023.
  
      Unusual venue: Scott Logic Lecture Theatre (MCS0001).
  
  Title: Large-scale to exascale data exploration and visualization with ParaView
  
  Abstract: After a decade of announcement, exascale computing is now a reality with the recent Frontier supercomputer starting up. Kitware has been deeply involved in the Exascale Computing Project (ECP) to participate in the development of new tools tailored for this major milestone. During this presentation, we will understand the challenges that large-scale simulations are facing regarding the exploration and visualization of their output, and how open-source tools like ParaView, Catalyst 2, VTK-m, ADIOS2, AMReX… would help to tackle these challenges.
  
  Short bio: François Mazen is the Assistant Director for the Scientific Visualization team at Kitware Europe. In 2008, François received his engineering degree at IFMA (French Institute for Advanced Mechanics) in Clermont-Ferrand where he was nominated for TOP 10 students. The same year, François also received a Master of Science at Université Blaise Pascal (Clermont-Ferrand) where he specialized in Rigid Body. His previous 13 years of experience, included 4 years at Ansys where he worked as a software developer in the Funded Development team, and more recently 6 years at Siemens PLM as Project Leader where he mainly worked on the design, architecture and development of Robot’s Path Planning Technology in C++ ans C#. With an extensive knowledge of project management, C++ development and visualization François strengthen KEU’s Scientific Visualization team proficiency.

• Wednesday, 10 May 2023, at 16:00, Garth Wells, University of Cambridge
  
  Title: Finite element methods at exascale
  
  Abstract: I will discuss the development of finite element algorithms and implementations for a range of applications on exascale hardware. When considering accelerators it is helpful to reflect on past attempts (since the honest efforts generally failed!) to assess why performance was disappointing. I will argue that the disappointing performance was due to a failure to assess the suitability of algorithms end-to-end; focusing on accelerating one step in a sequence of otherwise established of algorithms already over-constrained approaches and doomed them to failure. Improved mathematical and algorithmic understanding now allows us to exploit exascale-type hardware efficiently. Also on the upside, I will show that recent developments in compiler technologies have made it much easier to develop high performance finite element kernels without non-standard extensions, with measured performance compared against performance models. To make developments accessible, I will also touch upon the open-source FEniCS Project (https://fenicsproject.org) and its approach to high-level implementations with an assessment of the advantages and disadvantages of domain-specific languages and code generation for scientific computing. Code generation is no panacea! Finally, some recent performance data on the pre-exascale LUMI system will be presented.

• Wednesday, 26 April 2023, at 16:00, Benedict Rogers, The University of Manchester
  
  Title: Massive parallelisation of strictly incompressible flows using smoothed particle hydrodynamics
  
  Abstract: The meshless method, smoothed particle hydrodynamics (SPH) is becoming increasingly used in engineering industry for a range of applications such as aerospace, automotive, nuclear, chemical, offshore, marine, hydraulic and coastal engineering. The SPH method is now becoming competitive against well-established simulation techniques. Real problems are 3-D and require a massive number of particles and hardware acceleration, while the fluid itself can be considered a strictly incompressible. In this talk, we will consider how we have developed a massively parallel strictly incompressible SPH solver, the challenges involved, and how are now approaching this for the era of exascale computing.
  
  Short bio: Prof. Benedict D. Rogers is Chair of Computational Hydrodynamics and leads the Smoothed Particle Hydrodynamics (SPH) specialist group in the School of Engineering at the University of Manchester (UoM). He is a founder of the international organisation for SPH – the SPH rEsearch and engineeRing International Community (SPHERIC), and acted as its Chair (2015-2021). He is a core developer of the open-source code DualSPHysics, an international collaboration across 4 countries including the University of Parma and has been downloaded 100,000+ times to date. He has published over 70 peer-reviewed journal papers (H-index: 38) and is a co-Investigator on 2 projects preparing SPH for exascale computing. He has been awarded the Thomas Telford Premium Award by the Institution of Civil Engineers (ICE) twice – in 2014 and 2016 for work on SPH modelling of tsunami-structure interaction and in 2022 jointly received the prestigious International Joe Monaghan Prize for progress made addressing the Grand Challenges of SPH.

Epiphany term

• Thursday, 16 February 2023, at 15:00, Richard Graham, NVIDIA
  
      Part of the DPU Hackaton 2023.
  
  Title: NVIDIA’s BlueField DPU: offloading applications to the network
  
  Abstract: Plateauing of the capabilities of individual system components has led to new innovations in system design to meet growing computational needs. One such innovation is NVIDIA’s family of Data Processing Units (DPUs) which provides a network offload engine that include a Network Interface core, programmable CPU cores and targeted acceleration engines. This is known as the BlueField family of network adapters. This presentation will provide an overview of the BlueField network devices, principles behind making effective use of such devices and present work done using these devices to accelerate collective communication operations.

• Wednesday, 25 January 2023, at 16:00, Jemma Shipton, University of Exeter
  
  Title: Parallel timestepping algorithms for geophysical fluid dynamics
  
  Abstract: Following exciting developments in both mathematical analysis and practical experience, time-parallel methods are undergoing a revival as a potentially powerful route to exploiting future massively parallel exascale supercomputers. Time-parallel methods address the question of what to do when one has reached the limits of strong scaling (decreasing wallclock time by increasing the number of processors working in parallel) through domain decomposition parallelisation in space. A key lesson from the recent literature is that the success of parallel-in-time algorithms critically depends on them being carefully adapted to the equation being solved. Much like regular timestepping methods, there are many parallel-in-time algorithms, and the right algorithm needs to be designed and selected according to the mathematical properties and applications requirements of the underlying system. Here I will present an overview of several parallel-in-time algorithms that are relevant for weather and climate prediction, illustrating the theory in the context of ordinary differential equations and outlining future plans for applying them to geophysical flows.

Michaelmas term

• Wednesday, 23 Nov 2022, at 16:00, Rod Burns, Codeplay Software & Andrew Mallison, Intel
  
  Title: The SYCL GPU vision
  
  This talk is a follow-up to the group’s oneAPI workshop. While the workshop provides a platform for SYCL users, this talk targets a broader audience—those who might want to learn about SYCL from a high-level perspective. For in-person participants from Durham, the speakers will be available after the talk for 1:1 conversations around SYCL. These will be convered by an NDA.
  
  Abstract: The following will be covered:
  
  • oneAPI and SYCL and the heterogeneous landscape;
  • oneAPI on Intel;
  • oneAPI on Nvidia and AMD;
  • migrating CUDA to SYCL; and
  • oneAPI Community Forum collaboration.

• Wednesday, 16 Nov 2022, at 16:00, Jonas Latz, Heriot-Watt University
  
  Title: Gradient flows and randomised thresholding: sparse inversion and classification
  
  Abstract: Sparse inversion and classification problems are ubiquitous in modern data science and imaging. They are often formulated as non-smooth minimisation problems. In sparse inversion, we minimise, e.g., the sum of a data fidelity term and an L1/LASSO regulariser. In classification, we consider, e.g., the sum of a data fidelity term and a non-smooth Ginzburg–Landau energy. Standard (sub)gradient descent methods have shown to be inefficient when approaching such problems. Splitting techniques are much more useful: here, the target function is partitioned into a sum of two subtarget functions—each of which can be efficiently optimised. Splitting proceeds by performing optimisation steps alternately with respect to each of the two subtarget functions. In this work, we study splitting from a stochastic continuous-time perspective. Indeed, we define a differential inclusion that follows one of the two subtarget function’s negative subdifferential at each point in time. The choice of the subtarget function is controlled by a binary continuous-time Markov process. The resulting dynamical system is a stochastic approximation of the underlying subgradient flow. We investigate this stochastic approximation for an L1-regularised sparse inversion flow and for a discrete Allen–Cahn equation minimising a Ginzburg–Landau energy. In both cases, we study the longtime behaviour of the stochastic dynamical system and its ability to approximate the underlying subgradient flow at any accuracy. We illustrate our theoretical findings in a simple sparse estimation problem and also in low- and high-dimensional classification problems.

• Wednesday, 2 Nov 2022, at 16:00, Ana Lucia Varbanescu, University of Twente (and University of Amsterdam)
  
  Title: Towards zero-waste computing
  
  Abstract: “Computation” has become a massive part of our daily lives; even more so, in science, a lot of experiments and analysis rely on massive computation. Under the assumption that computation is cheap, and time-to-result is the only relevant metric for all of us, we currently
  use computational resources at record-low efficiency.
  In this talk, I argue this approach is an unacceptable waste of computing resources. I further define the goal of zero-waste computing and discuss how performance engineering methods and techniques can
  facilitate this goal. By means of several case-studies, I will also demonstrate performance engineering at work, proving how efficiency and time-to-result can co-exist.
  
  Short bio: Ana Lucia Varbanescu holds a BSc and MSc degree from POLITEHNICA University in Bucharest, Romania. She obtained her PhD from TUDelft, The Netherlands, and continued to work as a PostDoc researcher in The Netherlands, at TUDelft and VU University in Amsterdam. She is a MacGillavry fellow at University of Amsterdam, where she was tenured in 2018 as Associate Professor. Since 2022, she is also Professor at University of Twente. She has been a visiting researcher at IBM TJ Watson (2006, 2007), Barcelona Supercomputing Center (2007), NVIDIA (2009), and Imperial College of London (2013). She has received several NWO grants (including a Veni grant) and she is co-PI for the GraphMassivizer EU project.
  Ana’s research stems from HPC, and investigates the use of multi- and many-core architectures for HPC, with a special focus on performance and energy efficiency modeling for both scientific and irregular, data-intensive applications.
  
  A recording is available here.

• Wednesday, 26 Oct 2022, at 16:00, Katie Schuman, University of Tennessee
  
  Title: Opportunities for neuromorphic computing co-processors
  
  Abstract: Neuromorphic computing is a popular technology for the future of computing. Much of the focus in neuromorphic computing research and development has focused on new architectures, devices, and materials, rather than in the software, algorithms, and applications of these systems. In this talk, I will overview the field of neuromorphic computing with a particular focus on challenges and opportunities in using neuromorphic computers as co-processors. I will discuss neuromorphic applications for both machine learning and non-machine learning use cases.

• Wednesday, 12 Oct 2022, at 16:00, Nils Wedi, ECMWF
  
  Title: Destination Earth – digital twins of the Earth system
  
  Abstract: This talk will describe advances in the field of numerical weather prediction (NWP) and climate, culminating in the ongoing efforts to create digital replicas of the Earth system implemented by the European Commission’s Destination Earth initiative. Digital Twins of Earth encapsulate both the latest science and technology advances to provide near-real time information on extremes and climate change adaptation in a wider digital environment, where users can interact, modify and ultimately create their own tailored information.
  Recent work has demonstrated that global, coupled storm-resolving (or km-scale) simulations are feasible and can contribute to building such information systems and are no longer a dream thanks to recent advances in Earth system modelling, supercomputing and the ongoing adaptation of weather and climate codes for accelerators. Such simulations start to explicitly represent essential climate processes, e.g. detailed inland water and land-use representation, deep convection and mesoscale ocean eddies, that today need to be fully parametrised even at the highest resolution used in global weather and climate information production. These simulation outputs, combined with novel, data-driven deep learning advances, thus offer a window into the future, with a promise to significantly increase the realism and timeliness of delivery of Earth system information to a broad range of users. The significant compute and data challenges are discussed.

• Wednesday, 5 Oct 2022, at 16:00, Alex Titterton, Graphcore
  
  Title: Accelerating HPC workloads using AI and Graphcore’s IPU
  
  Abstract: For many years, researchers have been solving the world’s most complex scientific problems by undertaking traditional HPC techniques across a wide range of applications. Due to the growing complexity of calculations, operational costs, and the need to accelerate classical processes, an entirely new type of architecture is required. In this talk, we will provide a technical introduction to Graphcore’s Intelligence Processing Unit (IPU) and learn how traditional HPC workloads can be enhanced and accelerated using AI techniques running on the IPU. We’ll also explore how innovators are adopting this new approach across drug discovery, weather forecasting, climate modelling and computational fluid dynamics.

  Speaker bio: Alex Titterton has a PhD in particle physics, jointly awarded by the Universities of Bristol and Southampton, and has worked at CERN and the Rutherford Appleton Laboratory. Currently Alex works as Field Applications Engineer at Graphcore, where he has been enjoying tackling new challenges supporting Graphcore customers.

Academic year 2021/2022

• Wednesday, 18 May 2022, at 15:00, Fabian Knorr, University of Innsbruck,
  Title: Floating-point compressors

  Abstract: Storing and exchanging large amounts of floating-point data is common in distributed scientific computing applications. Data compression, when fast enough, can speed up such workloads by reducing contention on interconnects and storage systems. This talk explores two classes of floating-point compressors, the lossy and lossless type, and discusses their their utility on modern parallel and accelerator-based systems. We show which approach is best suited for what problem formulation and take a close look at ndzip, a lossless compressor for dense multi-dimensional data that is specifically engineered to achieve maximum efficiency on GPU-accelerated hardware.

  The Slides for this event are available here.

• Wednesday, 11 May 2022, at 15:00, Spencer Sherwin, Imperial College London, Title: Industry-Relevant implicit Large-Eddy Simulation of flows past automative and racing cars using  Spectral/hp Element Methods

  Abstract: We present the successful deployment of high-fidelity Large-Eddy Simulation (LES) technologies based on spectral/hp element methods to industrial flow problems that are characterized by high Reynolds numbers and complex geometries. In particular, we describe the steps required to perform the implicit LES of a realistic automotive and racing cars.  Notable developments had to be made in order to overcome obstacles in both mesh generation and solver technologies to simulate these flows, and will be outlined in this presentation. We thereby hope to demonstrate a viable pathway to translate academic developments into industrial tools, that can  advance the analysis and design capabilities of high-end engineering users.

• Wednesday, 4 May 2022, at 15:00, Joshua Short, Boston Limited, Title: NVIDIA and The Emergence of the Virtual World – Exploring The Omniverse, Digital Twins, and Cloud XR

  Abstract: With virtual reality becoming more integrated into the fabric of human existence, it’s evident that virtual existence is evolving and altering the social elements of life in new ways. But what benefits can virtual reality present to the world on a research or commercial level?
  We will be discussing how the Omniverse, Digital Twins, and Cloud XR technologies are being used in a variety of use cases, and how they can positively impact the future of not just virtual reality, but the physical world too.

  The Slides for this event are available here and here. A recording is available here.

• Wednesday, 27 Apr 2022, at 15:00, Carola Kruse, CERFACS, Title: On the efficient solution of saddle point systems with an inner-outer iterative solver based on the Golub-Kahan bidiagonalization

  Abstract: Symmetric indefinite linear systems of saddle point type arise in a large variety of applications, for example in fluid dynamics or structural mechanics. In this talk, we will review an inner-outer iterative solver for saddle point problems that is based on the Golub-Kahan bidiagonalization. The difficulty in the proposed solver is that in each outer loop an inner iterative system M, say, of size of the (1,1)-block has to be solved. If M arises from the discretization of a partial differential equation, efficient solvers might be available. Here, we will focus on the Stokes equation as a test problem and present different strategies for reducing the overall number of inner iterations and the computation time.

• Wednesday, 23 Mar 2022, at 15:00, Nils Deppe, California Institute of Technology & Lawrence Kidder, Cornell University, Title: SpECTRE: A task-based framework for astrophysics

  Abstract: Astrophysical phenomena vary greatly in spatial and temporal scales while also requiring complicated physics like neutrino transport. SpECTRE is a next-generation open-source (github.com/sxs-collaboration/spectre/) code designed with modern algorithms and computer science practices in order to take advantage of exascale supercomputers. We will discuss how SpECTRE uses the open-source task-based parallelization framework Charm++ (github.com/UIUC-PPL/charm/) to realize task-based parallelism. We will provide details on how our discontinuous Galerkin-finite-difference hybrid and elliptic solver algorithms are translated into the language of task-based parallelism, including the use of SIMD intrinsics and lazy evaluation. Time permitting, we will also provide an overview of our tensor library that allows scientists to write equations in a domain specific language, that of general relativity/gravity.

• Wednesday, 9 Mar 2022, at 15:00, Rosa Badia, Barcelona Supercomputing Center, Title: Parallel programming with PyCOMPSs and the dislib ML library

  Abstract: The seminar will present our group research on parallel programming models, more specifically in PyCOMPSs. PyCOMPSs is a task-based programming model for distributed computing platforms. In the seminar we will present the basics of the programming model and of its runtime. The seminar will also include some of our recent research work in the parallelization of machine learning with the dislib library. Dislib is a distributed, parallel machine learning library that offers a syntax inspired in scikit learn and it is parallelized with PyCOMPSs.

• Wednesday, 26 Jan 2022, at 15:00, Linus Seelinger, Heidelberg University, Title: Bridging the gap: Advanced uncertainty quantification and challenging models

  Abstract: Simulations of complex real-world processes, often by means of partial differential equations, lead to computational challenges. There is a rich ecosystem of methods and software packages to address these. The treatment of uncertainties in these models further increases problem dimensionality. However, the software ecosystem in the field of uncertainty quantification (UQ) is far less mature.
  This talk addresses the resulting gap between advanced UQ methods and challenging models in three ways:
  – An introduction to the MIT Uncertainty Quantification library (MUQ) is given. MUQ provides a modular framework for building UQ applications, offering numerous existing methods and reusable components.
  – A new universal interface for coupling UQ and model software is presented. Based on a simple HTTP protocol, it fully decouples development on both sides, while containerization provides portability.
  – A parallelized multilevel UQ method for high performance applications developed in MUQ is demonstrated at the example of a large-scale tsunami model.

• Wednesday, 12 Jan 2022, at 15:00, Johannes Doerfert, Argonne National Laboratory, Title: OpenMP in LLVM — Behind the Pragmas

  Abstract: OpenMP in LLVM is more than the directive handling in the frontends Clang and Flang. LLVM ships with various OpenMP specific compiler optimizations for a while now, and more are to come. There is a myriad of OpenMP runtimes to orchestrate portable accelerator offloading behind the scenes and to provide improved value beyond the OpenMP specification.
  In the talk we will explain how different implementation choices impact user experience and performance, either explicitly or due to their interaction with optimizations. In addition to best practices, participants will learn how to interact with the LLVM/Clang compiler to determine how OpenMP directives were implemented and optimized.
  Finally, we will give a brief overview of current developments in the LLVM/OpenMP world and how they can enable future exploration.

  The slides are available here.

• Wednesday, 17 Nov 2021, at 15:00, Hermann Haverkort, University of Bonn, Space-filling curves for tetrahedral meshes

  Abstract: With computations on adaptively refined meshes, one challenge is to achieve and maintain a good load balancing over multiple processors. A relatively simple and effective solution can be found in using a mesh that follows a fixed recursive tessellation, along with a space-filling curve that visits the tiles of this tessellation one by one. The decision how to cut the two- or higher-dimensional mesh into pieces now reduces to the much simpler decision where to cut the curve. If the curve is well-designed, contiguous sections of the curve are guaranteed to form well-shaped parts of the mesh, with well-shaped boundaries that enable efficient communication between the processors handling such sections.

  To generate and process meshes of triangles, squares, or cubes, there are a number of well-known space-filling curves with favourable properties. For meshes of tetrahedra or even higher-dimensional simplices the situation is more complicated: the question how to best generalise Polya’s triangle-filling curve to higher dimensions is still open. In this presentation I will present and propose several options, discuss their main properties and explain remaining challenges and open questions.

• Wednesday, 3 Nov 2021, at 15:00, Joseph Schuchart, University of Tennessee, Knoxville, Template Task Graphs for Irregular Task-based Applications

  Abstract: MPI and OpenMP still form the dominant programming paradigms for distributed and shared-memory programming and are commonly used in combination. However, more modern, C++ oriented approaches are gaining interest in the community. In this talk, I will present the Template Task Graphs (TTG), a C++-based approach that aims at providing a distribute task-based programming model that is both efficient and composable. By forming an abstract representation of the task-graph, TTG allows for the dynamic unrolling of task-graphs without prior knowledge of their exact shape and is thus especially suitable for irregular applications. I will present the current state of the model and first performance results on benchmarks resembling real-world target applications.

• Wednesday, 27 Oct 2021, at 15:00, Nicole Beisiegel, TU Dublin, An Adaptive Discontinuous Galerkin Model for Coastal Flood Simulations

  Abstract: Coastal flooding is an inherently complex phenomenon. This poses challenges for computer models with respect to computational efficiency, spatial resolution, or accuracy. In this talk, we will look at an adaptive discontinuous Galerkin (DG) model to simulate storm surge and coastal flooding more generally. A number of idealised testcases demonstrate the model’s performance. The adaptive, triangular mesh is driven by heuristic, or application-based refinement indicators. The discussion of the model’s computational efficiency will be guided by efficiency metrics that we define and apply to model results.

  This is joint work with J. Behrens (U Hamburg) and C.E. Castro (U Tarapaca).

• Wednesday, 6 Oct 2021, at 15:00, Edmond Chow, Georgia Tech, Introduction to Asynchronous Iterative Solvers

  Abstract: The standard iterative methods for solving linear and nonlinear systems of equations are all synchronous, meaning that in the parallel execution of these methods where some processors may complete an iteration before other processors (for example, due to load imbalance), the fastest processors must wait for the slowest processors before continuing to the next iteration.  This talk will discuss parallel iterative methods that operate asynchronously, meaning that the processors never wait for each other, but instead proceed using whatever iterate values are already available from other processors.  Processor idle time is thus eliminated, but questions arise about the convergence of these methods.  Asynchronous iterative methods will be introduced using simple fixed-point iterative methods for linear systems, before discussing asynchronous versions of rapidly converging methods, in particular, second order Richardson, and multigrid methods.

Academic year 2020/2021

• Friday, 27 Aug 2021, at 13:00, Adam Tuft, Durham University, A Tour of OMPT and Otter for Tracing Task-Centric OpenMP Programs

  Abstract: Reasoning about the structure of task-based programs, while vital for understanding their performance, is challenging for complex programs exhibiting irregular or nested tasking. The new OpenMP Tools (OMPT) interface defines event-driven callbacks allowing tools to gather rich runtime data on task creation and synchronisation. An example of such a tool is Otter (https://github.com/adamtuft/otter) which performs event-based tracing of OpenMP programs through OMPT, allowing the task-based structure of a target program to be recovered. This 30-minute presentation will give a brief tour of OMPT and will demonstrate its utility with examples provided by Otter.

• Thursday, 15 Jul 2021 09:30, Georg Hager, Performance counter analysis with Likwid and single node performance assessment

• Friday, Jun 25, 2021, at 13:00, Thomas Weinhart, University of Twente, Automated calibration for discrete particle simulations

  Abstract: The Discrete Particle Method (DPM) captures the collective behaviour of a granular material by simulating the kinematics of the individual grains. DPM can provide valuable insights that are difficult to obtain with either experiments or continuum models. However, calibrating the parameters of a DPM model against experimental measurements often takes significant effort and expertise, since automated and systematic calibration techniques are lacking.I will present an automated calibration technique, based on Bayesian filtering: We conduct experimental measurements to determine the material response, then simulate the same process in MercuryDPM, our open-source DPM solver [1], and measure the response of the simulated process. Then we apply a numerical optimisation algorithm to find the DPM parameters for which the response of the experiments and simulations match. This optimisation is done using a probabilistic optimisation technique called GrainLearning [2]. The technique can find local optima in only two to three iterations, even for complex contact models with many microscopic parameters.The technique has already been used in several projects, yielding good results. We present two test cases, one for calibrating a sintering model for 3D printing processes and one for calibrating the bulk response of a sheared granular material, and discuss the results.References:[1] Weinhart, T., Orefice, L., Post, M., et al, Fast, flexible particle simulations – An introduction to MercuryDPM, Computer Physics Communications, 249, 107129 (2020). [2] Cheng, H., Shuku, T., Thoeni, K. et al. Probabilistic calibration of discrete element simulations using the sequential quasi-Monte Carlo filter. Granular Matter 20, 11 (2018).

• Friday, 18 Jun 2021, at 13:00, Alexander Moskovsky, Moscow State University, RSC Group, Energy efficiency in HPC

  Abstract: High performance computing (HPC) is a pinnacle of modern computer engineering both regarding software and hardware components. Nowadays, they aggregate a large number of components: nodes, accelerators, storage devices and so on. One of the most acute problems on the hardware side is a rapid growth of energy dissipation and density: modern supercomputers consume megawatts of electric power. On the software side, the system software has to support the configuration of multiple components in concert.The RSC Group is a pioneer in liquid cooling for HPC solutions since the beginning of 2010. Liquid cooling enables 30-40% percent reduction of total supercomputer power consumption in comparison to forced airflow cooling. At the same time, liquid cooled HPC systems can be much more compact. RSC also develops a software stack that enables on-demand configurations for the both computational and storage systems, with support of Lustre and Intel DAOS and other filesystems. Such hyperconverged systems in HPC offer compactness and uniformity in hardware, but they require a software orchestrator to support it’s disaggregated architecture.RSC develops systems in close collaboration with it’s academic partners that inspire and motivate many solutions implemented in production. End-users from Russian Academy of Sciences, major Russian universities and research organisations tackle a wide spectrum of problems ranging from high energy physics to life sciences. RSC systems are present in supercomputer ratings like Top500, Green500, HPCG , IO500 and occupy 25% of Russian Top50 list of the most powerful computing systems.

• Friday, Jun 4, 2021, at 13:00, Benjamin Uekermann, University of Stuttgart, preCICE 2 – A Sustainable and User-Friendly Coupling Library

  Abstract: In the last five years, we have turned the coupling library preCICE from a working prototype to a sustainable and user-friendly community software project.In this presentation, I want to tell you about the challenges, success stories, and struggles of this endeavor, besides a brief introduction to the software itself. In particular, I cover documentation, tutorials, testing, integration with external simulation software, funding, and community building. Read more on https://precice.org/.

• Friday, 14 May 2021, at 13:00, Nicole Aretz, RWTH Aachen , Title: Sensor selection for linear Bayesian inverse problems with variable model configurations

  Abstract: In numerical simulations, mathematical models such as partial differential equations are widely used to predict the behavior of a physical system. The uncertainty in the prediction caused by unknown parameters can be decreased by incorporating measurement data: by means of Bayesian inversion a posterior probability distribution can be obtained that updates prior information on the uncertain parameters. As experimental data can be expensive, sensor positions need to be chosen carefully to obtain informative data despite a limited budget.In this talk we consider a group of forward models which are characterized through different configurations of the physical system. The configuration is a non-linear influence on the solution, e.g. the geometry or material of an individual work piece in a production chain. Our goal is to choose one set of sensors for the estimation of an uncertain linear influence whose measurement data is informative for all possible configurations. To this end, we identify an observability coefficient that links the experimental design to the covariance of the posterior. We then present a sequential sensor selection algorithm that improves the observability coefficient uniformly for all configurations. Computational feasibility is achieved through model order reduction. In particular, we discuss opportunities and challenges to decrease the computational cost of the inverse problem via the reduced basis method. We demonstrate our results on steady-state heat conduction problems for a thermal block and a geothermal model of the Perth Basin in Western Australia.

• Thursday, 8 Apr 2021, at 13:00, Jochim Protze, Title: Asynchronous MPI communication with OpenMP tasks

  Abstract: Your communication depends on computation results as input? Your computation task depends on data to arrive from a different process? OpenMP task dependencies should allow to express such dependencies. OpenMP 5.0 introduced detached tasks. In combination with MPI detached communication [1], this allows to build task dependency graphs across MPI processes. In this short presentation you will learn how you can integrate MPI detached communication into your project and profit from real asynchronous communication. If you don’t want to use OpenMP tasks, the same approach will also work with C++ futures/promises.

• Friday, 12 Mar 2021, at 13:00, Tim Dodwell, Alan Turing Institute, University of Exeter, Title: Adaptive Multilevel Delayed Acceptance

  Abstract: Uncertainty Quantification through Markov Chain Monte Carlo (MCMC) can be prohibitively expensive for target probability densities with expensive likelihood functions, for instance when the evaluation involves solving a Partial Differential Equation (PDE), as is the case in a wide range of engineering applications. Multilevel Delayed Acceptance (MLDA) with an Adaptive Error Model (AEM) is a novel approach, which alleviates this problem by exploiting a hierarchy of models, with increasing complexity and cost, and correcting the inexpensive models on-the-fly. The method has been integrated within the open-source probabilistic programming package PyMC3 and is available in the latest development version.

• Friday, 5 Feb 2021, at 13:00, Andy Davis, Courant Institute, Title: Super-parameterized numerical methods for the Boltzmann equation modeling Arctic sea ice dynamics

  Abstract: We devise a super-parameterized sea ice model that captures dynamics at multiple spatial and temporal scales. Arctic sea ice contains many ice floes—chunks of ice—whose macro-scale behavior is driven by oceanic/atmospheric currents and floe-floe interaction. There is no characteristic floe size and, therefore, accurately modeling sea ice dynamics requires a multi-scale approach. Our two-tiered model couples basin-scale conservation equations with small-scale particle methods. Unlike many other sea ice models, we do not average quantities of interest (e.g., mass/momentum) over a representative volume element. Instead, we explicitly model small-scale dynamics using the Boltzmann equation, which evolves a probability distribution over position and velocity. In practice, existing numerical methods approximating the Boltzmann equation are computationally intractable when modeling Arctic basin scale dynamics. Our approach decomposes the density function into a mass density that models how ice is distributed in the spatial domain and a velocity density that models the small-scale variation in velocity at a given location. The mass density and macro-scale expected velocity evolve according to a hyperbolic conservation equation. However, the flux term depends on expectations with respect to the velocity density at each spatial point. We, therefore, use particle methods to simulate the conditional density at key locations. We make each particle method independent using a local change of variables that defines micro-scale coordinates. We model small-scale ice dynamics (e.g., collision) in this transformed domain.