Department of Mathematics

CoSy Seminars Autumn 2015

1 September
Speaker: Martin Gaskell, Dept. Astronomy & Astrophysics, University of California at Santa Cruz
Title: Understanding what happens as  supermassive black holes are fed
Place: Ångström 11167
Time:12.00-13.00
Local contact person: Beatriz Villarroel Rodriguez

8 September
Speaker: Matteo Magnani, Computing Science, IT Department, Uppsala University
Title: Multilayer Social Network Analysis: state of the art and future trends
Place: Ångström 11167
Time:12.00-13.00
Local contact person: Mike Ashcroft

15 September
Speaker: Raaz Sainudiin, School of Mathematics and Statistics, University of Canterbury, Christchurch, New Zealand
Title: Mayonnaise, Mustard and Mathematics: a hydrodynamic limit of a correlated site percolation model of a viscoplastic fluid
Place: Ångström 11167
Time:12.00-13.00
Local contact person: Warwick Tucker

22 September
Speaker: Venelin Mitov, Dept. of Biosystems Science and Engineering, ETH, Zürich, Switzerland
Title: Importance of pathogen mutation in quantifying the viral contribution to virulence of an HIV infection
Place: Ångström 11167
Time:12.00-13.00
Local contact person: Krzysztof Bartoszek

29 September
Speaker:  Oleg Kochukhov, Dept. of Physics and Astronomy, Uppsala University
Title: Cartography of stellar surfaces
Place: Ångström 11167
Time:12.00-13.00

13 October
Speaker: Ali Syed, Founder and Chief Data Scientist of Persontyle
Title: The Age of Data Driven Science and Engineering
Place: Ångström 11167
Time:12.00-13.00
Local contact person: Mike Ashcroft

3 November
Speaker: Mike Ashcroft, Computing Science, IT Department, Uppsala University
Title:Multiclass Delaunay Field Estimation & Effective Identification of High Value Areas of the Feature Space
Place: Ångström 11167
Time:12.00-13.00
Local contact person: Mike Ashcroft

17 November
Speaker: Jesse Silverberg, Wyss Institute for Biologically Inspired Engineering, Harvard University, Boston, MA, USA
Title: Experiments at the Intersection of Geometry, Mechanics, and Microstructure
Place: Ångström 11167
Time:12.00-13.00
Local contact person: Arianna Bottinelli

24 November
Speaker: Gregory Feiden, Dept. of Physics and Astronomy, Uppsala University
Title: Probing the Properties of Stellar Convection with Statistical Inference
Place: Ångström 11167
Time:12.00-13.00

1 December
Speaker: Carlos Castillo
Title: TBA
Place: Ångström 12167
Time:12.00-13.00
Local contact person: Mike Ashcroft

7 December
Speaker: Olivier Gagliardini, Laboratoire de Glaciologie et Géophysique de l'Environnement, University Joseph Fourier, Paris, France
Title: Friction, basal hydrology and their interactions : a modelling perspective
Place: Ångström 11167
Time:12.00-13.00
Local contact person: Per Lötstedt

8 December
Speaker: Erik Zackrisson, Dept. of Physics and Astronomy, Uppsala University
Title: Machine learning in the study of the first galaxies
Place: Ångström 11167
Time:12.00-13.00

15 December
Speaker: Shunsuke Muto, Institute of Materials and Systems for Sustenability, Nagoya University
Title: Mind the Noise: Mining Hidden Information from Spectroscopic Datasets
Place: Ångström 12167
Time:12.00-13.00

1st September , 2015
Speaker: Martin Gaskell, Dept. Astronomy & Astrophysics,University of California at Santa Cruz
Title: UNDERSTANDING WHAT HAPPENS AS SUPERMASSIVE BLACK HOLES ARE FED
Local contact person: Beatriz Villarroel Rodriguez
Abstract: What would it be like to be next to a supermassive black hole that is swallowing gas?  Rapid feeding of a supermassive black hole can produce the most powerful continuous sources of energy in the universe, active galactic nuclei (AGNs).  In this talk I describe the recent major paradigm changes my collaborators and I have proposed for our understanding of conditions and processes close to an actively accreting supermassive black hole.  I argued that these paradigm changes solve major problems and that a coherent picture of the AGN phenomenon is emerging.

Date: 8 September, 2015
Speaker: Matteo Magnani, Computing Science, IT Department, Uppsala University
Title: Multilayer Social Network Analysis: state of the art and future trends
Abstract:  In this talk I will present the multilayer approach to studying social networks, which has recently become one of the main research topics in the area of network science. With the aim of interesting a heterogeneous audience, I will provide a broad overview of the area, highlighting the differences and opportunities with respect to traditional research in social network analysis and mining. Network co-evolution, clustering of interdependent networks, multi-dimensional information paths and network transformation & visualization will be among the concepts touched upon in my overview, all rigorously equation-free, to hopefully provide a pool of discussion topics for the subsequent fika.

Date: 15 September, 2015  
Speaker: 
Raaz Sainudiin, Senior Lecturer in the School of Mathematics and Statistics, University of Canterbury, Christchurch, New Zealand.
Title: Mayonnaise, Mustard and Mathematics: a hydrodynamic limit of a correlated site percolation model of a viscoplastic fluid
Abstract: We present a Gibbs random field model for the microscopic interactions in a viscoplastic fluid. The energy function is derived from the Gibbs potential in terms of the external stress and internal energy. The resulting Gibbs distribution, over a configuration space of microscopic interactions, can mimic experimentally observed macroscopic behavioral phenomena that depend on the externally applied stress. A simulation algorithm that can be used to approximate samples from the Gibbs distribution is given and it is used to gain several insights about the model. The model has two parameters for the internal energy of the material in the absence of external stress and a third parameter for a constant externally applied stress. An approximating differential equation for the expected proportion of the material in the solid phase is derived by a spatio-temporal rescaling of the toroidal square lattice upon which the Gibbs random field model is defined. The asymptotic dynamics of this tri-parametric family of differential equation matches with those of the rescaled simulations from the Gibbs field model and can account for the macroscopic behaviors, including solid-fluid phase transitions in the presence of constant as well as varying external stress and the associated hysteresis.  The model matches with data from shearing mayonnaise and mustard.
References:
http://dx.doi.org/10.1039/c5sm00857c
http://tinyurl.com/q589hfy

Date: Sep. 22nd
Speaker: 
Mitov Venelin, Department of Biosystems Science & Engineering, Mattenstrasse 26, 4058 Basel, Switzerland
Title: Importance of pathogen mutation in quantifying the viral contribution to virulence of an HIV infection
Abstract: The viral load set point (SPVL) is the best known predictor of the virulence of an HIV-1 infection. The relative contributions of viral genotype and patient’s immune system in determining SPVL have been controversially discussed. The heritability, quantifying the importance of the viral genotype, was estimated between 5.7% and 52% (1-3). This difference in an order of magnitude has been explained by differences of the cohorts under study as well as differences in the estimation methods. Some studies were based on regression of recipient- on donor SPVL values in known transmission couples, e.g. (3), while other studies borrowed phylogenetic methods from macroevolution such as the Pagel’s lambda (2) and the Phylogenetic Mixed Model (PMM) (1).
Recently, simulation studies revealed that the phylogeny-based methods might not provide accurate estimates (4). We argue that donor-recipient regression methods are also prone to systematic bias because these methods ignore within-host mutation of the pathogen. We present a new tool to estimate the heritability of a trait, based on an extension of the PMM method assuming an Ornstein-Uhlenbeck (OU) instead of a Brownian Motion (BM) model of evolution of the pathogen contribution to the phenotype of an infection. Thus, our method takes into account within-host mutation as well as possible selection on the trait of interest. Applying the method to data from Switzerland and the UK, we observe that the HIV genotype has a significant impact on virulence, with estimated heritability above 0.2.
References:
1.     E. Hodcroft et al., The Contribution of Viral Genotype to Plasma Viral Set-Point in HIV Infection. PLoS Pathog. 10, e1004112 (2014).
2.     S. Alizon et al., Phylogenetic approach reveals that virus genotype largely determines HIV set-point viral load. PLoS Pathog. 6, e1001123 (2010).
3.     T. D. Hollingsworth et al., HIV-1 transmitting couples have similar viral load set-points in Rakai, Uganda. PLoS Pathog. 6, e1000876 (2010).
4.     G. Shirreff et al., How effectively can HIV phylogenies be used to measure heritability? Evolution, Medicine, and Public Health. 2013, 209–224 (2013).

Date: Sempteber 29, 2015
Speaker: Oleg Kochukhov, Associate Professor, Division of Astronomy and Space Physics, Department of Physics and Astronomy,Uppsala University
Title: Cartography of stellar surfaces
Abstract: With a few exceptions, stars are too far away to allow a direct study of their surfaces, even using the largest telescopes available on Earth. Yet, from the example of the Sun, we know that characterising dynamic surface behaviour (spot formation, emergence and evolution of magnetic fields, pulsations) is critical for understanding the physics of stars and their impact on circumstellar environment, including planetary systems. It turns out that by studying profiles of the absorption lines visible in stellar spectra it is possible to detect spots and magnetic fields. Moreover, by applying a sophisticated mathematical inversion procedure one can reconstruct two-dimensional scalar or vector maps of stellar surfaces with a spatial resolution by far exceeding direct imaging capabilities of world’s largest telescopes. In this seminar I outline the basic principles and present examples of the application of this powerful astrophysical remote sensing method.

Date: 13 October
Speaker: Ali Syed, Founder and Chief Data Scientist of Persontyle
Title: The Age of Data Driven Science and Engineering
Abstract: The speaker is founder and chief data scientist of one of the companies involved in the European Data Science Academy, a project funded by the European Union horizon 2020 funding program. He will explain the idea behind the European Data Science Academy, how these ideas are being implemented and what the future holds.

Date: 3 November
Speaker: Mike Ashcroft, Computing Science, IT Department, Uppsala University
Title: Multiclass Delaunay Field Estimation & Effective Identification of High Value Areas of the Feature Space
Abstract: I will explain how we can use the Delaunay tessellation of sample data to reconstruct the density field of the underlying distribution. This reconstruction can then be used for efficient discovery of high value regions in the feature space. Such models and these queries are both highly scalable both regarding the number of sample points and the number of dimensions. The talk will give an overview of the origin of the technique proposed.  I will discuss natural neighbor interpolation techniques based on the Voronoi tessellations, as well as point density estimation techniques using both tessellations. I think these are particularly interesting as techniques that are relatively unknown within the data science/machine learning community, despite being used in other areas of science and technology – including single class density estimation in far space astrophysics.I will introduce the R package I have developed for working with these techniques and discuss future work on testing the relative performance of the proposed techniques. I will also briefly discuss the potential application of the tools developed within image recognition.

Date: 17 November
Speaker: Jesse Silverberg, Wyss Institute for Biologically Inspired Engineering, Harvard University, Boston, MA, USA
Title: Experiments at the Intersection of Geometry, Mechanics, and Microstructure
Place: Ångström 11167
Time:12.00-13.00
Local contact person: Arianna Bottinelli
Abstract:Climbing cucumbers, popping pollen grains, wrinkled fingers, and curly hair.  At heart, the modern revival of mechanics covers a diverse range of subjects at the intersection of function and form.  It's at this point, where geometry, mechanics and microstructure meet, that we find buckling instabilities, mechanical phase transitions, exotic stress responses, and fracture.  While these phenomena are widely observed in many inert materials, we also find them being actively employed in biological tissues, where they have evolved as essential tools for survival.
In this talk, I'll address two specific tissues and discuss experiments that initially focus on mechanics, but naturally lead to questions of microstructure and geometry.  The first project explores plant root growth, an essential topic as agricultural crop demands continue to increase with the rising global population.  Using 3D time-lapse imaging, we uncover a novel strategy that utilizes passive mechanical instabilities to navigate barriers in the growth environment.  Exploring this phenomenon further, we also find a growth strategy for seeking the steepest downward direction that has features remarkably similar to E. coli’s well-known run-and-tumble chemotactic behavior.
The second project explores articular cartilage, a biological material that enables smooth and painless joint motion.  Using a combination of experimental techniques, an unusual structure-function relationship for this material is empirically determined, and a model based on percolating fiber networks is offered as a solution. 
In the end, a central theme will emerge: mechanical instabilities are not simply a problem to be avoided, but rather, they're an opportunity to be leveraged for functional and responsive materials.

24 November
Speaker: Gregory Feiden, Dept. of Physics and Astronomy, Uppsala University
Title: Probing the Properties of Stellar Convection with Statistical Inference
Place: Ångström 11167
Time:12.00-13.00
Abstract
A majority of stars in the Galaxy are low-mass main-sequence stars. These stars are actively fusing hydrogen in their core and have an outer convection zone. Convection is an important physical process that regulates the transport of energy within low-mass stars, controlling the whole of a star's interior structure and many of a star's observable properties. Briefly, I will review the basic physics of stellar convection along with a simple phenomenological description used in stellar interior structure models. Under the simplified model of convection, convective properties are typically assumed to be qualitatively similar to properties of convection in the Sun within some scaling factor. I'll present results from ongoing investigations that leverage high precision data and statistical inference to provide empirical constraints on the properties of convection in other stars within the framework of stellar structure models. Particular attention will be given to the statistical techniques and the robustness of the results.

7 December
Speaker: Olivier Gagliardini, Laboratoire de Glaciologie et Géophysique de l'Environnement, University Joseph Fourier, Paris, France
Title: Friction, basal hydrology and their interactions : a modelling perspective
Place: Ångström 11167
Time:12.00-13.00
Local contact person: Per Lötstedt
Abstract:
During the last decades, Greenland surface melt has shown an increase both in intensity and spatial extent. Part of this water probably reaches the bedrock and enhances the glacier speed, advecting larger volume of ice into the ablation area. In the context of a warming climate, this mechanism will contribute to the future rate of retreat and thinning of the land-terminating glaciers of Greenland. Complex couplings, implying both positive and negative feedbacks, prevail between surface mass balance, ice flow, basal hydrology and the evolution of the glacier geometry. Larger amount of melt water might increase or decrease the mean ice flow of a glacier, depending on the capacity of the basal hydrology network to evacuate this surplus of water, which in turn will influence the surface crevassing and the ability of water to reach the bedrock at higher elevations. 
In this presentation, I will first present our current knowledge about the representation of basal friction in ice flow models, its dependency to water pressure and how the spatial distribution and time evolution of the hydrological system can be modelled. I will then present a newly developed framework in the open source, finite element, ice sheet / ice flow model Elmer/Ice which allow to solve the prognostic equations for ice flow and the hydrology model in a fully coupled manner. I will finally come back on the Greenland example and show that the currently observed crevasse distribution is likely to extend upstream, leading to an increase in water reaching the bedrock, and in turn, accelerates the retreat and thinning of land-terminating glaciers.

8 December
Speaker: Erik Zackrisson & Hannes Jensen , Dept. of Physics and Astronomy, Uppsala University
Title: Machine learning in the study of the first galaxies
Place: Ångström 11167
Time:12.00-13.00
Abstract:
The most distant galaxies we know of today appear just as tiny specks of light in the deepest astronomical images obtained by the Hubble Space Telescope. Since the photons from these galaxies have travelled for more than 13 billion years to reach us, we see these objects not as they are today, but as they were a few hundred million years after the Big Bang. By studying objects at such distances, astronomers are hoping to find clues to some of the unsolved puzzles of early Universe: How did supermassive black holes form? What was the nature of the first stars? What caused the reionization of the Universe? The launch of the James Webb Space Telescope in 2018 will revolutionize this field by providing the first detailed light spectra of some these earliest generations of galaxies. In this seminar, we will present an interdisciplinary project aiming to make the most of the upcoming James Webb Space Telescope data, by applying machine learning techniques to the analysis of simulated galaxy spectra.

15 December
Speaker: Shunsuke Muto, Institute of Materials and Systems for Sustenability, Nagoya University
Title: Mind the Noise: Mining Hidden Information from Spectroscopic Datasets
​Place: Ångström 12167
Time:12.00-13.00
Abstract:
Introduction
The digital technologies have changed current scientific measurements to automated operations, which has improved their efficiency and accuracy particularly in many repeated measurement processes, thereby inevitably enlarging the data size obtained. Such ‘big data’ obtained without arbitrary choices of specific areas of interest may contain richer information than those obtained with specific purposes/expectations. It is now sophisticated methods called as ‘data mining’ that are available for mining embedded information from the datasets based on information/statistics theories.
Spectrum imaging (SI) techniques by scanning a small probe of electron/x-ray/light on a sample area of interest can provide such a dataset as the two-dimensional data array, each row corresponding to the spectrum at a specific position. A non-negativity matrix factorization (NMF) technique [1] [2] can then apply to the dataset, which decomposes the set of spectra into a product of the constituent pure spectral components and their corresponding relative composition (weight) matrices without any reference spectra. This method allows us to provide a two dimensional spatial distribution map of different chemical states incorporated even when the multiply overlapped spectra.
In the present talk we introduce our concept of the NMF technique, particularly applied to datasets obtained by scanning transmission electron microscopy (STEM) and electron energy-loss spectroscopy (EELS), followed by discussing its intrinsic difficulties and our attempts to solve them.
Formalism
The present NMF algorithm is based on the multiplicative update rules, with both the Euclid square distance and the generalized Kullback-Leibler divergence applied to minimize the distance between the experimental data and the derived solution for modelling the Gaussian and Poisson noise, respectively [2]. Critical issues in the method are (i) how one can determine the number of pure spectral components involved and (ii) that the solution is not unique. We imposed the automatic relevance determination (ARD) condition for (i) [3] and the orthogonality constraint on the composition matrix for (ii) [4].
Model data
A STEM-EELS-SI dataset obtained from a cross-section sample of a semiconductor memory device was used for the test data in order to confirm the relevance of the method. It was found that a slight inclusion of the orthogonal constraint effectively suppressed cross-talks between the pure spectral components, otherwise revealed. The effect of ARD will be also discussed.
References
[1] S. Muto, T. Yoshida and K. Tatsumi, Mater. Trans. 50, 964 (2009).
[2] H. Sawada, J. Inst. Electr. Info. Comm. Eng. 95, 829 (2012) (in Japanese).
[3] V. Y. F. Tan and C. Fevotte, IEEE Trans. Pat. Anal. Mach. Intell. 35, 1592 (2013).
[4] K. Kimura, Y. Tanaka and M. Kudo, Proc. ACML 2014 39, 129 (2014).