16 January 2024

Andrea Beck, Stuttgart: Scientific Machine Learning for Flow Simulations – Successes, failures and a way ahead

10:30-11:30 in MH:Riesz

Fika will be served from 10:00 for those who registered. The seminar can be attended without registration.

Description

Andrea Beck is a full professor for Numerical Methods in Fluid Mechanics and deputy director of the institute for aerodynamics and gas dynamics at the University of Stuttgart, Germany.

In this talk, I will give an overview of scientific machine learning applied to fluid mechanics, in particular to the numerical simulation of flows and flow-related problems. My main focus will be the combination of scientific computing and machine learning, combining mathematical models with data-driven decision making and high performance computing. I will show some successful and also not so successful examples from my own journey into augmenting PDE solvers with ML models - from turbulence simulation to shock capturing and flow control. Along the way, I will discuss lessons learned and conclude by providing my perspective on the still missing ingredients towards true scientific machine learning for flow simulations.

30 November 2022

Göran Johansson, Chalmers: Quantum Computing - Hardware development and the search for use-cases in Sweden

Abstract

In this talk, I’ll give a brief overview of the ongoing efforts to build a superconducting quantum computer in the Wallenberg Center for Quantum Technology (WACQT). Together with our industry partners, we are also investigating possible use-cases and I will briefly describe this activity as well. In particular I will mention logistics optimisation, quantum chemistry and protein folding.

15 February 2022

Frank Krauss: Introducing JUNE - an open-source epidemiological simulation

Abstract

We constructed a detailed digital twin of the UK population, with supreme social and geographical granularity, representing 55 million residents in England and Wales and tracing their daily movements and activities. The infection is propagated through the virtual population through simulated social contacts, and the progress of the disease in infected individuals and their trajectory through the healthcare system is then simulated, based on public health data. We resolve the spatio-temporal development of the disease spread through the population in great detail and are able to find non-trivial correlations of infection and fatality rares with a variety of societal factors. Our model has been proven in the midst of the current crisis and is currently being used by NHS-England’s COVID-19 response team to inform their strategic and operational planning.

Speaker presentation

Information about the speaker at the Institute for Particle Physics Phenomenology, Durham University, UK 

1 December 2021

Pauline Kergus - Physics-informed learning for identification of a residential building's thermal behavior

This was a joint AI Lund/COMPUTE seminar.

Speaker presentation

Pauline Kergus uses artificial intelligence to model the thermal behavior of buildings. As space heating represents a large share of total energy use, thermal networks, i.e. district cooling or heating networks, would be able to increase the efficiency of the energy system in an economic way. Thanks to the natural inertia of heat exchanges, these networks can offer flexibility. In order to explore this feature, it is important to model building's thermal behavior in order to enable the use of demand-side management control strategies. Pauline's research focuses on building such models through a physics-informed learning based approach, taking advantage of the available measurements. The goal is to use the obtained models to control power consumption and to build predictive models for production planning.

More information about Pauline Kergus at the website of Automatic Control

16 November 2021

David Sumpter: Why complexity science failed and what we can do to save it

Speaker presentation

David Sumpter is a professor of applied mathematics at Uppsala University. He studies the mathematics of collective animal behaviour, social dynamics and mathematical biology.  David is the author of Soccermatics and Outnumbered, which have been translated into ten languages, and Collective Animal Behaviour, the leading text in the academic field he helped create. He has worked with a number of the world's biggest football clubs, advising on analytics,

21 October 2021

Fabian Theis: Single-cell latent space learning across labs, modalities and perturbations

This seminar was part of the Swedish Bioinformatics Workshop.

Speaker presentation

Fabian Theis uses artificial intelligence to unlock the secrets of human cells. Single cell sequencing combined with deep learning enables him to analyse and model differences between cells. Fabian is the director of the Institute of Computational Biology at the Helmholtz Center Munich and scientific director of the Helmholtz Artificial Intelligence Cooperation Unit (HelmholtzAI) which was launched in 2019. He is a full professor at the Technical University of Munich, holding the chair ‘Mathematical Modelling of Biological Systems’, associate faculty at the Wellcome Trust Sanger Institute as well as adjunct faculty at the Northwestern University.

20 October 2021

Oliver Stegle: From genotype to phenotype with single-cell resolution

This seminar was part of the Swedish Bioinformatics Workshop.

Speaker presentation

Oliver Stegle is the Head of the Computational Genomics and Systems Genetics Division at the German Cancer Research Center (DKFZ) and group leader at EMBL in Heidelberg, Germany. He uses computational methods to unravel the genotype–phenotype map on a genome-wide scale.

His research team carries out work at the interface of statistical inference, machine learning and computational biology, pioneering computational methods for integrating large and heterogeneous datasets across individual and at the single-cell level.

19 October 2021

From Robots to Rust: Artificial Intelligence and Sustainable Development

Abstract

AI is an innovative technology that will greatly influence the world's development. But how can we balance its benefits and dark sides to support sustainable development? AI is more than a model taking decisions, it includes a long chain of people, tasks and resources we must consider. As part of LU's Future Week, COMPUTE, AI Lund and the Agenda 2030 Graduate school will host a public seminar at which we will discuss the relation between sustainability and artificial intelligence. We will show examples of how AI can be a tool for sustainability projects and also discuss the importance and potential ways of making AI-based research and supercomputing more sustainable. We invite users, developers and other computing experts to discuss with us.

Programme

• 17:00-17:20: "AI as a tool and target for sustainability projects"
  Sonja Aits, Associate Senior Lecturer, Faculty of Medicine, Lund University
• 17:20-17:30: "Working with sustainability in supercomputing facilities"
  Anders Follin, Technical Director, LUNARC, Lund University
• 17:30-18:00: Panel discussion: "AI and sustainability - challenges and opportunities"
  Panel members: Sonja Aits, Anders Follin, Erik Wilson (Project Manager, AI Sweden), Markus Grillitsch (Director of CIRCLE, Lund University)
  Moderator: Juan Ocampo, PhD student, Agenda 2030 Graduate School

14 April 2021

Jan Malmgreen: Veberöd - the research village for a digital and sustainable society

This was a joint AI Lund/COMPUTE seminar.

Abstract

Veberöd and it surrounding villages are Sweden's first test bed for smart villages ("Smarta Byar"). This allows researchers, students, companies and other partners to conduct projects that can contribute to a sustainable and digital development of the country side.

Projects can make use of the 3D model of Veberöd, to which live data can be connected, and a platform for communication with the residents. There are also opportunities to test IoT solutions and collaborate with local companies.

Link

Home page of Smarta Byar (in Swedish).

7 April 2021

Mattias Ohlsson: Information-driven care the CAISR Health initiative

This was a joint AI Lund/COMPUTE seminar.

Abstract

The availability of data is changing rapidly in healthcare. Information-driven care will make use of all this data, together with data analytics and machine learning, with the aim improve the healthcare system. The CAISR Health research profile at Halmstad University will focus on information-driven care, understanding the whole chain from formulating and prioritising questions, to algorithms, to data collection, to engagement, to explainability, and to implementation. 

Mattias Ohlsson will provide an overview of CAISR Health and briefly present some applications.

22 March 2021

Sarah Gibson - The Turing Way: Reproducible research and beyond!

Abstract

Reproducible research is necessary to ensure that scientific work can be trusted. Funders and stakeholders are beginning to require that publications and research outreach include access to the underlying data and analysis code. The goal is to ensure that all results can be independently verified and built upon in future work. This is sometimes easier said than done! Sharing these research outputs means understanding data management, library sciences, software development, and continuous integration techniques. The Turing Way is a handbook to support research professionals, stakeholders, funders, students and their supervisors in ensuring that reproducible research is "too easy not to do". It includes training material on version control, analysis testing, and open and transparent communication with future users. Beyond just the reproducibility aspect of research, The Turing Way has expanded into a series of volumes sharing insights on building data science projects that are open, collaborative, inclusive and ethical. This project is openly developed and any and all questions, comments and recommendations are welcome at our GitHub repository: https://github.com/alan-turing-institute/the-turing-way.

3 March 2021

Eran Elhaik (Molecular Cell Biology, Lund University): Answering the big WHEN and WHERE questions in paleogenomics with machine learning

This was a joint AI Lund/COMPUTE seminar.

Abstract

Ancient DNA (aDNA) has changed the studies of history, enabling us to directly analyse early genetic variation. In recent years, there has been a sharp increase in the amount of collected aDNA and high-profile studies, but insufficient information about timing and geographical origin has limited the usefulness of the collected data and resulted in many erroneous reports. To correctly interpret paleogenomic data, it is important to answer two basic questions: WHEN and WHERE does the genetic material come from?

The WHEN question is usually answered with radiometric dating (RD). But, half of all published aDNAs are missing reliable and accurate RD.

The WHEN question is usually answered by assuming that the burial site is the same as the original habitat. All in all, this results in many vague and contradictory reports and erroneous conclusions.

These difficulties hinder studies of the origins of humans and domesticated animals.

We developed a machine learning-based approach to overcome these problems. In this talk, I will share the origin of Eurasian populations in minute details as emerged from our work.

19 October 2020

K. J. Åström (Automatic Control, LTH): Modelling and Simulation - from Physics to Languages and Software

Abstract

The lecture gives a broad overview of modelling and simulation from the perspective of computing and control. Vannevar Bush built the mechanical differential analyser to simulate power systems. Analogue computing developed rapidly because of the advances in electronics. It was widely used for simulation both in industry and academia by the 1950s. The initial efforts to exploit digital computing for simulation were based on emulation of analogue computers. The tedious effort to go from physics to a working analogue simulation was simplified significantly by introducing modelling languages. Languages were first developed for specific areas and later extended to broader physical domains. Hilding Elmqvist developed the language Dymola for multi-domain physical modelling in his 1978 PhD dissertation at the Department of Automatic Control at LTH. The key idea was to cut a system into subsystem and use object orientation to structure the subsystems. The behaviour of the subsystems was expressed by physics in its basic form of balance and constitutive equations. Symbolic and numeric computing was used to transform the model to code suitable for optimization and simulation. Toyota’s use of Dymola in the development of Toyota Prius in 1996 was a major industrial breakthrough. The formation of Modelica Association (Modelica.org), a nonprofit organization for open standards and software for physical modelling, was a major advance. Modelica is now widely used in automotive, aerospace and other industries. Funding for Modelica projects in the period 2007-2022 is more than 100 M€. There are regular Modelica Conferences since 1996. In 2020 there are conferences both in Japan and the USA. There are regular design meetings since 1996, the 100th meeting was held in Lund in 1919. There are three companies in Lund devoted to Modelica, Dassault Systèmes (formerly Dynasim), Modelon and Mogram.

Link

A video of the seminar is available at YouTube.

14 September 2020

Daniel Tamayo (Princeton): Generalization and confidence intervals for machine learning models in the Sciences: Application to predicting the long-term stability of planetary systems

Abstract

Several hundred multi-planet systems have been discovered around stars other than our own. These are indirect and challenging detections at the limit of observational capabilities, which often leave many important physical parameters uncertain, notably masses and orbital eccentricities. Interestingly, many of these multi-planet systems are in compact orbital configurations, which in principle allow one to rule out the wide ranges of parameter space that quickly lead to dynamical chaos and planetary collisions. Unfortunately, direct evaluation of stability through numerical integration is computationally prohibitive due to the long dynamical ages of these worlds.

This unsolved problem of determining the long-term stability of given orbital configurations has a rich history, partially motivating the discovery of dynamical chaos and driving the development of non-linear dynamics. I will present our latest work combining our partial understanding of the dynamics in these systems with machine learning methods to significantly improve on previous efforts and computationally open up of the stability constrained characterization of exoplanet systems. Our complementary work on this problem highlights the tradeoffs between manual feature engineering and data-driven features discovered through deep learning for complex problems in the sciences. Finally, I will discuss our latest efforts to go beyond simple point-estimates to training models that additionally provide confidence intervals using Bayesian Neural Networks and the newly developed Multi-SWAG algorithm (Wilson & Izmailov 2020).

Link

A video of the seminar is available at YouTube.

21 April 2020

Jens Saak (Max Planck, Magdeburg): Fighting the Reproducibility Crisis - Sustainable research software and RRR for computer-based experiments

Abstract

Computer-based experiments have gained increasing importance over the last few decades. Nonetheless, sufficient documentation of how computational results have been obtained in experiments described in research publications is often not available. In some cases, even the replication of ones own results is difficult for researchers. The reproduction of other people’s findings on their own computational setup is a daily challenge for researchers around the world. The reusability in new contexts to accelerate the evolution of knowledge at the rate one should be able to expect is often entirely out of scope.

In the recent few years this has been observed as an issue across the sciences and the term “reproducibility crisis” has evolved. The ultimate goal of easy reuse of software developed for a research project by a completely different group in a potentially very different computational environment is likely to far fetched, since in the end researchers are typically not software engineers. This is due to the lack of software development education outside computer science, and would require a whole new infrastructure with positions for research software engineers, having enough knowledge about the field to understand the methodology, but also being educated in modern software development strategies. Still, in a work- group or a research community the reusability-goal can be addressed.

We discuss simple recommendations that make replicability, reproducibility and reusability easier to achieve. Furthermore, we provide basic guidelines for the handover of research software projects that increase their lifetime, make their development more sustainable and enable researchers to find answers to their scientific challenges rather than reinventing the wheel writing yet another basic code for their communities favorite academic toy problem.

Links

The slides are available for download (PDF, 4 MB, new tab).

A video of the seminar is also available (MP4, 96 MB, new tab).

24 January 2020

Carolina Wählby (Uppsala University): Developing computational approaches in microscopy

18 November 2019

Enrico Ronchi (Fire Safety Engineering, LTH): Modelling crowd evacuation

6 May 2019

Maggie Lieu (European Space Agency): Detecting solar system objects with convolutional neural networks

Joint COMPUTE/AIML seminar

Abstract

With upcoming big astronomical surveys such as LSST and Euclid imaging enormous volumes of the sky at high speeds, the detection of asteroids, comets and other moving bodies with traditional methods will not be feasible. Machine learning methods such as convolutional neural networks can help us to sift through the large amounts of data and quickly identify objects of interest for follow up observation and tracking. In this talk I will show how CNNs can be trained to effectively identify asteroids and other astronomical objects in an unbiased manner even when the training data sample is small.

26 March 2019

Costas Bekas (IBM Research, Zurich): Cognitive Discovery - How AI is Changing Technical R&D

Joint COMPUTE/AIML event

Abstract

Cognitive Discovery (CD) is an AI-based holistic framework designed to massively accelerate technical R&D. It introduces novel AI technologies to ingest and represent complicated technical knowledge and know-how, sift through oceans of data and drive technical hypothesis discovery and verification by clever design of simulations. Real market deployments of CD have demonstrated at least an order of magnitude improvement in discovery rates of new technologies such as chemicals and materials, while in addition driving the creation and curation of corporate deep technical memory.

Speaker biography

Dr. Costas Bekas is managing the Foundations of Cognitive Solutions group at IBM Research-Zurich. He received B. Eng., Msc and PhD diplomas, all from the Computer Engineering & Informatics Department, University of Patras, Greece, in 1998, 2001 and 2003 respectively. Between 2003-2005, he worked as a postdoctoral associate with prof. Yousef Saad at the Computer Science & Engineering Department, University of Minnesota, USA. He has been with IBM since September 2005. Costas's main research interests span AI, massive scale analytics and energy aware algorithms and architectures. Costas is a recipient of the PRACE 2012 award and the ACM Gordon Bell 2013 and 2015 prizes.

March 18 2019

Andrew Winters (Mathematical Institute, University of Cologne): Stability in High-Order Numerics

Abstract

Nature is non-linear. Many fundamental physical principles such as conservation of mass, momentum, and energy are mathematically modeled by non-linear time dependent partial differential equations (PDEs). Such non-linear conservation laws describe a broad range of applications in science and engineering, e.g., the prediction of noise and drag from aircraft, the build up and propagation of tsunamis in oceans, the behavior of gas clouds, and propagation of non-linear acoustic waves in different materials.

In practice, marginally resolved simulations of non-linear conservation laws, such as the compressible Navier-Stokes equations or the ideal magnetohydrodynamics (MHD) equations, reveal that high-order methods are prone to aliasing instabilities. This can lead to total failure and breakdown of the algorithms. Aliasing issues are introduced and intensified by a combination of insufficient discrete integration precision, collocation of non-linear terms, polynomial approximations applied to rational functions, and, again, by insufficient grid resolution. The aim of this talk is to discuss a remedy for such aliasing issues and present its connection to discrete variants of the product and chain rules. As such, we construct nodal high-order numerical methods for non-linear conservation laws that are entropy stable. To do so, we focus on particular approximate derivative operators used to mimic steps from the continuous well-posedness PDE analysis on the discrete level and enhance the robustness of the high-order numerics.

6 December 2018

Benjamin Ragan-Kelley (Simula Research Laboratory): Jupyter: facilitating interactive, open, and reproducible science

16 November 2018

Björn Nystedt (NBIS): Presentation of NBIS — National Bioinformatics Infrastructure Sweden

21 September 2018

Steven Longmore (Liverpool John Moores University): Using machine learning to identify animals from drones

Link

A video of the seminar is available at YouTube.

Abstract

The World Wildlife Fund for Nature (WWF) estimates that up to five species of life on our planet become extinct every day. This astonishing rate of decline has potentially catastrophic consequences, not just for the ecosystems where the species are lost, but also for the world economy and planet as a whole. Indeed, biodiversity loss and consequent ecosystem collapse is commonly listed as one of the 10 foremost dangers facing humanity, and most pressingly in the developing world. There is a fundamental need to routinely monitor animal populations over much of the globe so that conservation strategies can be optimized with such information. The challenge faced to meet this need is considerable. To date most monitoring of animal populations is conducted manually, which is extremely labour-intensive, inherently slow and costly. Building on technological and software innovations in astronomy and machine learning, we have developed a drone plus thermal infrared imaging system and an associated automated detection/identification pipeline that has the potential to provide a cost-effective and efficient way to overcome this challenge. I will describe the current status of the system and our efforts to enable local communities in developing countries with little/no technical background to run routine monitoring and management of animal populations over large and inhospitable areas and thereby tackle global biodiversity loss.

8 May 2018

Alice Quillen (University of Rochester): Astro-viscoelastodynamics or Soft Astronomy: Tidal encounters, tidal evolution and spin dynamics

Abstract

Mass spring models, originally developed for graphics and gaming applications, can measure remarkably small deformations while conserving angular momentum. By combining a mass spring model with an N-body simulation, we simulate tidal spin down of a viscoelastic moon, directly tying simulated rheology to orbital drift and internal heat generation. I describe a series of applications of mass spring models in planetary science. Close tidal encounters among large planetesimals and moons were more common than impacts. Tidal encounters can induce sufficient stress on the surface to cause large scale brittle failure of an icy crust. Strong tidal encounters may be responsible for the formation of long chasmata in ancient terrain of icy moons such as Dione and Charon. The new Horizons mission discovered that Pluto and Charon's minor satellites Styx, Nix, Kerberos, and Hydra, are rapidly spinning, but surprisingly they have spin axis tilted into the orbital plane (they have high obliquities). Simulations of the minor satellites in a drifting Pluto-Charon binary system exhibit rich resonant spin dynamics, including spin-orbit resonance capture, tumbling resonance and spin-binary resonances. We have found a type of spin-precession mean-motion resonance with Charon that can lift obliquities of the minor satellites in the Pluto/Charon system.

12 April 2018

Paul-Christian Bürkner (University of Münster): Why not to be afraid of priors (too much)

Link

The talk was given at Bayes@Lund 2018. Videos and slides from the event are available at GitHub.

26 March 2018

Martin Turbet (Laboratoire de Météorologie Dynamique, Paris): Exploring the diversity of planetary atmospheres with Global Climate Models

Abstract

More than 50 years ago, scientists created the first Global Climate Models (GCMs) to study the atmosphere of the Earth. Since then, the complexity and the level of realism of these models (that can now include the effect of oceans, clouds, aerosols, chemistry, vegetation, etc.) have considerably increased. The large success of these models have recently motivated the development of an entire family of GCMs designed to study extra-terrestrial environments in our solar system (Venus, Mars, Titan, Pluto) and even beyond (extrasolar planets).

I will first show various GCM applications on Venus, Mars, Titan and Pluto. Solar system GCMs successes and sometimes failures teach us useful lessons to investigate and possibly predict the possible climates on planets where no (or almost no) observations are available. I will then present several examples of studies recently performed using a 'generic' Global Climate Model developed at the Laboratoire de Météorologie Dynamique in Paris, designed to explore the possible atmospheres and the habitability of ancient planets or extrasolar planets.

1 December 2017

Bojana Rosić (TU Braunschweig): Uncertainty quantification in a Bayesian setting

Abstract

Concrete and human bone tissue are typical examples of materials which exhibit randomness in the mechanical response due to an uncertain heterogeneous micro-structure. In order to develop an appropriate probabilistic macro-scale mathematical description, the essential step is to address the material as well as possible other sources of uncertainties (e.g. excitations, change in geometry , etc.) in the model. By extending already existing deterministic models derived from Helmholtz free energy and the dissipation functions characterising ductile or quasi-brittle behaviour, the goal of this talk is to identify and quantify uncertainty in the system response. For this purpose a Bayesian probabilistic setting is considered in which the modeller's a priori knowledge about the model parameters and the available set of data obtained by experiments are taken into account when identifying the corresponding probability distribution functions of unknown parameters. Identification in the form of Bayesian inverse problems - in particular when experiments are performed repeatedly - requires an effcient solution and representation of possibly high dimensional probabilistic forward problems, i.e. the estimation of the measurement prediction given prior assumption. An emergent idea is to propagate parameter uncertainties through the model in a Galerkin manner in which the solution of the corresponding differential equations is represented by a set of stochastic basis polynomials, the cardinality of which grows exponentially. To allow an effcient solution of high-dimensional problems this talk will present the new low-rank Galerkin schemes combined with Bayesian machine learning approaches.

30 May 2017

Michel Defrise (Vrije Universiteit, Brussel): Statistical reconstruction methods in medical imaging

Abstract

Statistical reconstruction methods are probably today the most common methods to obtain tomographic images from SPECT/PET measurements. The methods are based on modelling the camera systems and thereby iteratively find a good estimate of the internal radionuclide distribution from the similarity of the calculated and measured projection data. The main advantage is here that if physical problems, associated with the measurement of the radiation, such as, non-homogeneous photon attenuation, contribution from scattered photons, partial volume problems due to collimator resolution, septal penetration etc, can be modelled accurately then this naturally comes out as a compensation for the problems. The lecture will cover the fundamental mathematical parts of the procedures describe above.

14 November 2016

INTEGRATE meeting - See the INTEGRATE website for more details.

26 October 2016

INTEGRATE math-hack day - See the INTEGRATE website for more details.

5 October 2016

INTEGRATE hack day - See the INTEGRATE website for more details.

26 September 2016

INTEGRATE meeting - See the INTEGRATE website for more details.

11 May 2016

INTEGRATE meeting - See the INTEGRATE website for more details.

25 April 2016

INTEGRATE math-hack day - See the INTEGRATE website for more details.

12 April 2016

Patrick Farrell (Oxford University): Automated adjoint simulations with FEniCS and dolfin-adjoint

Abstract

The derivatives of PDE models are key ingredients in many important algorithms of computational science. They find applications in diverse areas such as sensitivity analysis, PDE-constrained optimisation, continuation and bifurcation analysis, error estimation, and generalised stability theory.

These derivatives, computed using the so-called tangent linear and adjoint models, have made an enormous impact in certain scientific fields (such as aeronautics, meteorology, and oceanography). However, their use in other areas has been hampered by the great practical difficulty of the derivation and implementation of tangent linear and adjoint models. Naumann (2011) describes the problem of the robust automated derivation of parallel tangent linear and adjoint models as ``one of the great open problems in the field of high-performance scientific computing''.

11 April 2016

INTEGRATE hack day - See the INTEGRATE website for more details.

14 March

INTEGRATE meeting - See the INTEGRATE website for more details.

12 October 2015

Volker Springel (Heidelberg Institute for Theoretical Studies): Cosmic structure formation on a moving mesh

Abstract

Recent years have seen impressive progress towards hydrodynamic cosmological simulations of galaxy formation that try to account for much of the relevant physics in a realistic fashion. At the same time, numerical uncertainties and scaling limitations in the available simulation codes have been recognized as important challenges. I will review the state of the field in this area, highlighting a number of recent results obtained with large particle-based and mesh based simulations. I will in particularly describe a novel moving-mesh methodology for gas dynamics in which a fully dynamic and adaptive Voronoi tessellation is used to formulate a finite volume discretization of hydrodynamics which offers numerous advantages compared with traditional techniques. The new approach is fully Galilei-invariant and gives much smaller advection errors than ordinary Eulerian codes, while at the same time offering comparable accuracy for treating shocks and an improved treatment of contact discontinuities. The scheme adjusts its spatial resolution to the local clustering of the flow automatically and continuously, and hence retains a principle advantage of SPH for simulations of cosmological structure growth. Applications of the method in large production calculations that aim to produce disc galaxies similar to the Milky Way will be discussed.

16 April 2015

Derek Richardson (University of Maryland): Asteroids: Modeling the future of space exploration

Abstract

Over the past 2 decades since the first spacecraft images were returned from an asteroid, we have learned that these potentially hazardous objects not only often support satellites but are likely themselves loose collections of fragmented material. Numerical simulations assist in understanding the evolution of these leftover building blocks of the terrestrial planets, planning for the mitigation of their threat to Earth, and for advising space mission designers on the challenges of sample return and possible astronaut landing. Our group uses an adapted high-performance parallelized gravity tree code (PKDGRAV) to simulate gravitational and collisional processes among small bodies in space. Basic features of the code will be presented, along with their application to selected topics in small-body evolution. The talk will feature our latest simulations of granular flow in microgravity, a key tool for modeling sample-return mechanisms.

23 March 2015

Thomas Neuhaus (Jülich Supercomputing Centre): Quantum computing via quantum annealing from a physics point of view

Abstract

I present an overview on the theory of quantum annealing for the use of solving so called intractable mathematical problems. I will introduce satisfiability problems consisting of a number of simple constraints (clauses) on a set of Boolean variables (e.g., 2SAT and 3SAT) and report on today's knowledge on the usefulness of quantum annealing in such theories. I will also give an introduction to the existing Dwave Quantum Computer and present a few computer experiments on the machine. As of today theoretical physics has not yet identified a class of mathematical problems, nor physical theories, that really benefits from quantum annealing with certainty. A search for such problems is under way.

16 February 2015

Gregor Gassner (University of Cologne): A massively parallel model for fluid dynamics simulations

Abstract

In this talk, I will consider the numerical simulation of non-linear advection-diffusion problems, particularly of the compressible Navier-Stokes equations used to model turbulent fluid flows in engineering applications and problems appearing in natural sciences.

For this, we develop stable and accurate methods for conservation laws and incorporate other important physical aspects, such as e.g. entropy stability. Examples include the Burgers equation, the shallow water equations, Maxwell's equations and the Euler equations.

A special emphasis is on the computational implementation of these mathematical models such that they are well suited for realistic applications. Due to the multi-scale character of the considered problems, a large amount of spatial and temporal resolution is needed for an accurate approximation. The number of spatial degrees of freedom are as high as one billion with over one hundred thousand time steps. Such simulations are only feasible, when the power of today's largest supercomputers are unleashed.

15 December 2014

Anna-Karin Tornberg (KTH): Accelerated boundary integral simulations of particulate and two-phase flows

Abstract

In micro-fluidic applications where the scales are small and viscous effects dominant, the Stokes equations are often applicable. The suspension dynamics of fluids with immersed rigid particles and fibers are very complex also in this Stokesian regime, and surface tension effects are strongly pronounced at interfaces of immiscible fluids - such as surfaces of water drops in oil.

Simulation methods can be developed based on boundary integral equations, which leads to discretizations of the boundaries of the domain only, and hence fewer unknowns compared to a discretization of the PDE. This involves evaluating integrals containing the fundamental solution (Green's function) for the PDE. This will result in both singular and nearly singular integrals that need to be evaluated, and the construction of accurate quadrature methods is a main challenge. The Green's functions decay slowly, which results in dense or full system matrices. To reduce the cost of the solution of the linear system, an acceleration method must be used. If these two issues - accurate quadrature methods and acceleration of the solution of the linear system - are properly addressed, boundary integral based simulations can be both highly accurate and very efficient.

17 November 2014

Debora Sijacki (University of Cambridge): Simulating galaxy formation: numerical and physical uncertainties

Abstract

Hydrodynamical cosmological simulations are one of the most powerful tools to study the formation and evolution of galaxies in the fully non-linear regime. Despite several recent successes in simulating Milky Way look-alikes, self-consistent, ab-initio models are still a long way off. In this talk I will review numerical and physical uncertainties plaguing current state-of-the-art cosmological simulations of galaxy formation. I will then present global properties of galaxies as obtained with novel cosmological simulations with the moving mesh code Arepo and discuss which physical mechanisms are needed to reproduce realistic galaxy morphologies in the present day Universe.

20 October 2014

Lund University Humanities Lab — Why the humanities needs a lab and what we do init

Speakers

• Marianne Gullberg, Director of the Humanities Lab: Why on earth do the Humanities need a lab?
• Annika Andersson: What the recording of brainwaves can tell us about language processing
• Marcus Nyström: What are eye movements? Looking into the elastic eye
• Victoria Johansson: Are we reading during writing? What does the combination of keystroke logging and eyetracking reveal?
• Nicolo dell'Unto: The use of 3D models for intra-site investigation in Archaeology

Abstract

Lund University Humanities Lab is an interdisciplinary research and training facility whose aim is to enable scholars (mainly) in the humanities to combine traditional and novel methods, and to interact with other disciplines in order to meet the scientific challenges ahead. The Lab hosts technology, methodological know-how, and archiving expertise, and a wide range of research projects. Activities are centered around issues of communication, culture, cognition and learning, but many projects are interdisciplinary and conducted in collaboration with the social sciences, medicine, the natural sciences, engineering, and e-Science locally (Lund University), nationally, and internationally. We start with a brief overview of the lab and its facilities, followed by four short talks exemplifying the kind of research that takes place in the Lab.

Link

Find out more at the web site of Lund University Humanities Lab.

27 May 2014

Freddy Ståhlberg (Dept. of Medical Radiation Physics): What is LBIC?

Abstract

The overall goal of Lund University Bioimaging Center (LBIC) is to pursue in-depth knowledge of human metabolism and function by developing and combining advanced imaging techniques, primarily high-field MRI, PET and SPECT). The goal is accomplished by a step-by-step establishment of the above mentioned bio-imaging center at our University, using already well-established research groups in the field as the core structure.

To achieve this goal, we provide technical platforms, knowledge and support in the development of new diagnostic methods from experimental models, while at the same time taking unmet clinical problems to device patient-specific tailored therapies. Technically, MRI and PET will be advanced both as individual modalities and also towards a merged use. We envision future simultaneous extraction of physiological and molecular information in a multimodal imaging environment for better understanding the impact of molecular events on cellular/tissue behavior. We believe that the development of the Lund University Bioimaging Center will have a significant impact on the research at Lund University. Locally, the center is a powerful resource for translational research at the medical faculty and at Skåne university hospital, with strong connections to planned larger research facilities in Lund, e.g. MAXIV and ESS Scandinavia. On a national and international level, the buildup of the proposed center can be foreseen to become an asset to the whole biomedical research community.

5 May 2014

Martin Rosvall (Umeå University): Memory in network flows and its effects on community detection, ranking, and spreading

Abstract

It is a paradigm to capture the spread of information and disease with random flow on networks. This conventional approach ignores an important feature of the dynamics: where flow moves to depends on where it comes from. That is, memory matters. We analyze multi-step pathways from different systems and show that ignoring memory has profound consequences for community detection. Compared to analysis without memory, community detection with memory generally reveals system organizations with more and smaller modules that overlap to a greater extent. For example, we show that including memory reveals actual travel patterns in air traffic and multidisciplinary journals in scientific communication. These results suggest that we can use only more data and not more elaborate algorithms to identify real modules in integrated systems.

23 April 2014

Paul Segars (Duke University): The XCAT phantom based on NURBS to create realistic anatomical computer models for use in Medical Imaging simulations

Talks on research involving voxel-based phantoms at Dept of Medical Radiation Physics, Lund University:

• Michael Ljungberg: Nuclear medicine and the use of voxel-phantoms in Monte Carlo research
• Gustav Brolin: A national quality assurance study of dynamic 99Tcm-MAG3 renography using Monte Carlo simulations and the XCAT phantom
• Katarina Sjögreen Gleisner: An introduction to image-based dosimetry
• Johan Gustafsson: A procedure for evaluation of accuracy in Image-Based dosimetrical caculations using the XCAT Phantom

9 December 2013

Thomas Curt (Irstea): Modelling fire behaviour and simulating fire-landscape relationships: Possibilities and Challenges

Abstract

Fire is one of the major disturbances on the global scale, shaping landscapes and vegetation, and affecting the global carbon cycle (Gill et al., 2013). Global changes (land use changes and climate change) are predicted to modify fire regimes and fire distribution in many parts of the world (Krawchuk et al., 2009). In this context assessing fire risk and simulating fire-landscape interactions become crucial to sustainable land management. In the recent period, important progress has been made to model fire behavior and to simulate fires on various spatial scales. A variety of models and simulators now exist, which have specific abilities and limitations (see www.fire.org/). We propose to review some of these models, and to sum up their purpose, potential, and challenges. Fire behavior models (e.g. BehavePlus, Firetec) allow simulating fire behavior for almost all ecosystems using quite simple field data. They also permit to assess fire effects on the ecosystems, notably postfire tree mortality. Some of these models are fully physics-based (2D or 3D), while others use empirical equations. They have been used or adapted to a large array of ecosystems and climate conditions worldwide, and examples will be presented. Landscape-fire models (e.g. Farsite, FlamMap) are spatially-explicit and designed to simulate fires with different weather scenarios in sufficiently realistic landscapes (Cary et al., 2006). Thus they consider the combined effects of weather, fuels, and topography. Recently, high performance fire simulation systems have been developed to permit the simulation of hundreds of thousands of fires. They provides large-scale maps of burn probability or flame length, thus allowing to assess which ecosystems and areas are at risk and should be monitored and managed preferentially. Some applications based on this type of models will be presented. Coupling fire behavior models with vegetation into Dynamic Global Vegetation Model (e.g. LPJ, Sierra) allows studying the role of fire disturbance for global vegetation dynamics. Their application at regional scale requires an approach which is explicit enough to simulate geographical patterns, but general enough to be applicable to each vegetation type at large scales. We shall present how the Sierra model can be used at local scale, in order to predict vegetation shifts according to different scenarios of future fire regime.

27 May 2013

Kevin Heng (University of Bern): Exoplanetary Atmospheres and Climates: Theory and Simulation

Abstract

Exoplanet discovery is now an established enterprise. The next frontier is the in-depth characterization of the atmospheres and interiors of exoplanets, in preparation for next-generation observatories such as CHEOPS, TESS and JWST. In this presentation, I will review the astronomical, astrophysical and computational aspects of this nascent field of exoplanet science. I will begin by reviewing the observations of transit (absorption) and eclipse (emission) spectra and phase curves, emphasizing the importance of understanding clouds/hazes. Next, I will review the astrophysical aspects of understanding exoplanetary atmospheres, including the need to elucidate the complex interplay between dynamics, radiation and magnetic fields. Finally, I will discuss the Exoclimes Simulation Platform (ESP), an open-source set of simulational tools currently being constructed by my Exoplanets and Exoclimes Group at the University of Bern, reviewing the technical challenges faced by the ESP team. Advancing the theoretical state of the art requires a hierarchical (1D models versus 3D simulations) and multi-disciplinary approach, drawing from astronomy, astrophysics, geophysics, atmospheric and climate science, applied mathematics and planetary science.

7 May 2013

Susanna C. Manrubia (Centro de Astrobiología, Madrid): Tinkering in an RNA world and the origins of life

Abstract

Life arose on Earth some 3.8 billion years ago. Despite tremendous experimental and theoretical efforts in the last 60 years to uncover how the first protocells could have originated from abiotic matter, we are still facing many more unknowns than answers. Two main players in the process are self-replicating molecules and metabolic cycles, whose emergence requires the presence of chemical species with the ability to code genetic information and catalyze chemical reactions. In current living organisms, DNA performs the former function and proteins the latter. But there is one molecule, RNA, able to perform both functions. This fact has led to the hypothesis that an ancient RNA world could have preceded modern cellular organization. That attractive scenario opens the possibility of carrying out computational studies on issues such as how RNA molecules could have been selected for simple chemical functions, what is the probability that such a molecule would arise in prebiotic environments in the absence of faithful template replication, or how populations of molecules explore the space of sequences and disclose evolutionary innovations.

10 December 2012

Bernhard Mehlig (Göteborg University): Turbulent aerosols: clustering, caustics, and collisions

Abstract

Turbulent aerosols (particles suspended in a turbulent fluid or a random flow) are fundamental to understanding chemical and kinetic processes in many areas in the Natural Sciences, and in technology. Examples are the problem of rain initiation in turbulent clouds, and the problem of describing the collisions and aggregation of dust grains in circumstellar accretion disks.

In this talk I summarise recent progress in our understanding of the dynamics of turbulent aerosols. I discuss how the suspended particles may cluster together, and describe their collision dynamics in terms of singularities in the particle motion (so-called caustics).

25 October 2012

Chris Lintott (Oxford University): What to do with 600,000 scientists

Abstract

'Citizen science' - the involvement of hundreds of thousands of internet-surfing volunteers in the scientific process - is a radical solution to the problem of Big Data. The largest and most successful such project, Galaxy Zoo, has seen volunteers provided more than 200 million classifications of more than 1 million galaxies, going beyond simple classification task to make and even follow-up serendipitous discoveries of their own. As PI of Zooniverse.org, astronomer Chris Lintott leads a team that has enabled volunteers to sort through a million galaxies, discover planets of their own, determine whether whales have accents and even transcribe ancient papyri. In this talk, he will review some of the highlights of their work, explore the technology needed to keep such a large group of volunteers busy and look forward to a future when man and machine can work in harmony once more...

7 June 2012

Cristian Micheletti (International School for Advanced Studies, Trieste): Coarse-grained simulations of DNA in confined geometries.

Abstract

Biopolymers in vivo are typically subject to spatial restraints, either as a result of molecular crowding in the cellular medium or of direct spatial confinement. DNA in living organisms provides a prototypical example of a confined biopolymer. Confinement prompts a number of biophysics questions. For instance, how can the high level of packing be compatible with the necessity to access and process the genomic material? What mechanisms can be adopted in vivo to avoid the excessive geometrical and topological entanglement of dense phases of biopolymers? These and other fundamental questions have been addressed in recent years by both experimental and theoretical means. Based on the recent reviews of refs. [1,2] we shall given an overview of these studies and report on general simulation techniques that can be effectively used to characterize the equilibrium properties of coarse-grained models of confined biomolecules.

References

[1] D. Marenduzzo et al. J. Phys.: Condens. Matter 22 (2010) 283102: Link to the article at IOP Science

[2] C. Micheletti et al. Physics Reports, 504 (2011) 1-73: Link to the article at Science Direct (may require subscription)

Link

The slides are available for download (PDF, 3.5 MB, new tab).