Past Events

Seminars are held Tuesdays and Thursdays in POB 6.304 from 3:30-5:00 pm, unless otherwise noted. Speakers include scientists, researchers, visiting scholars, potential faculty, and ICES/UT Faculty or staff. Everyone is welcome to attend. Refreshments are served at 3:15 pm.

Thursday, Nov 16

Nonparametric Bayesian data analysis

Thursday, Nov 16, 2017 from 3:30PM to 5PM | POB 6.304

  • Additional Information

    Hosted by Tan Bui-Thanh

    Sponsor: ICES Seminar

    Speaker: Peter Mueller

    Speaker Affiliation: Professor, UT Austin

  • Abstract

    We review inference under models with nonparametric Bayesian (BNP) priors. The discussion follows a set of examples for some common inference problems. The examples are chosen to highlight problems that are challenging for standard parametric inference. We discuss inference for density estimation, clustering, regression and for mixed effects models with random effects distributions. While we focus on arguing for the need for the flexibility of BNP models, we also review some of the more commonly used BNP models, thus hopefully answering a bit of both questions, why and how to use BNP.


  • Additional Information

    Hosted by Ron Elber

    Sponsor: ICES Seminar-Molecular Biophysics Series

    Speaker: Scott Showalter

    Speaker Affiliation: Pennsylvania State University

  • Abstract

    Intrinsically Disordered Proteins (IDPs) partially or completely lack a co-operatively folded structure under native conditions, making their equilibrium state very different from that typically described through high-resolution structural biology. Our view is that IDPs do possess native structure that is responsible for imparting their specific functions; describing these structures simply requires a broadening of the traditionally narrow structure-function paradigm. To better understand the function of IDPs, our laboratory focuses on transcription factors and the enzymes that carry out transcription in eukaryotes. In this presentation, we will focus on the Carboxy-Terminal Domain (CTD) of the RNA polymerase II (Pol II) large subunit in order to illustrate our approach. CTD cycles through multiple phosphorylation states that correlate with progression through the transcription cycle and regulate nascent mRNA processing. Structural analyses of yeast and mammalian CTD have been hampered by their repetitive sequences. Here we identify a region of the Drosophila melanogaster CTD that is essential for Pol II function in vivo and capitalize on natural sequence variations within it to facilitate structural analysis. Mass spectrometry and NMR spectroscopy reveal that hyper-Ser5 phosphorylation transforms the local structure of this essential region via proline isomerization. The sequence context of this switch tunes the apparent activity of the CTD phosphatase Ssu72, suggesting a mechanism for the selective recruitment of cis-proline specific regulatory factors that may synergize with CTD phosphorylation to augment gene regulation in developmentally complex organisms.


Friday, Nov 10

  • Additional Information

    Hosted by Federico Fuentes and Sriram Nagaraj

    Sponsor: ICES Seminar-Babuska Forum Series

    Speaker: Amir Shahmoradi

    Speaker Affiliation: ICES, Dept. of Aerospace Engineering and Engineering Mechanics

  • Abstract

    Model inadequacy and measurement uncertainty are two of the most confounding aspects of inference and prediction in quantitative sciences. The process of scientific inference (inverse problem) and prediction (forward problem) involve multiple steps of data analysis, hypothesis formation, model construction, parameter estimation, model validation, and finally prediction of the quantity of interest. This talk seeks to clarify the concepts of model inadequacy, model bias, and measurement uncertainty, and the two traditional classes of uncertainty: aleatoric vs. epistemic, as well as their relationships with each other. Starting from basic principles of probability, I build and explain a hierarchical Bayesian framework to quantitatively deal with model inadequacy and noise in data. I explain how this general approach can resolve many existing logical paradoxes that frequently appear in
    experimental data. As an illustrative problem, I apply this methodology to an in-vitro dataset of the growth of liver tumor cells subject to significant imaging background noise. I explain how this general approach can retrieve the unknown quantities of interest from the available noisy data without any logical inconsistencies, such as obtaining negative values for quantities that are known to be inherently positive. Finally, I discuss some recent developments in the field of stochastic integration techniques with applications to numerical computation of Bayesian evidence.

    Bio:
    Dr. Amir Shahmoradi holds a PhD in Physics from the University of Texas at Austin with special focus on Biophysical sciences and a Master of Science in the field of High Energy Astrophysics. He is currently a Peter O'Donnell fellow at Institute for Computational Engineering and Sciences, and lecturer at the Department of Aerospace Engineering and Engineering Mechanics at the University of Texas at Austin. His current research is in the area of computational oncology, Bayesian Data Analysis methods, and stochastic optimization and integration techniques.


Thursday, Nov 9

Fabrication-Aware Geometry Processing

Thursday, Nov 9, 2017 from 3:30PM to 5PM | POB 6.304

  • Additional Information

    Hosted by Tom Hughes

    Sponsor: ICES Seminar

    Speaker: Daniele Panozzo

    Speaker Affiliation: Assistant Professor, Courant Institute of Mathematical Sciences, New York University

  • Abstract

    The advent of commodity 3D printing is revolutionizing the way people think about designing and prototyping: a designer can now hold in her hands a 3D object hours after its design is complete, drastically reducing costs and enabling quick iterations over many designs. However, the majority of software tools and algorithms currently used to create, manipulate, and process digital geometry are not fabrication-aware: they model the shape as an abstract entity that often does not satisfy practical requirements such as stability or robustness. This leads to a large gap between the digital design and the physical fabrication — my research strives to fill this gap, providing computational design tools that rely on numerical optimization to create fabrication-ready designs. In this talk, I will present recent results on the design of masonry and tensegrity structures and show how computation can be used to add textures to complex geometric shapes using hydrographic printing and thermoforming techniques.

    Bio:
    Daniele Panozzo is an Assistant Professor of Computer Science at the Courant Institute of Mathematical Sciences in New York University. Prior to joining NYU he was a postdoctoral researcher at ETH Zurich (2012-2015). He earned his PhD in Computer Science from the University of Genova (2012) and his doctoral thesis received the EUROGRAPHICS Award for Best PhD Thesis (2013). Daniele’s research interests are in digital fabrication, geometry processing, architectural geometry and discrete differential geometry. He received the EUROGRAPHICS Young Researcher Award in 2015, the NSF CAREER Award in 2017, and his work has been covered by Swiss National Television and various national and international printed media. Daniele is leading the development of libigl (https://github.com/libigl/libigl), an award-winning (EUROGRAPHICS Symposium of Geometry Processing Software Award, 2015) open-source geometry processing library that supports academic and industrial research and practice. Daniele is chairing the Graphics Replicability Stamp (http://www.replicabilitystamp.org), which is an initiative to promote reproducibility of research results and to allow scientists and practitioners to immediately benefit from state-of-the-art research results.


Tuesday, Nov 7

Geometric electromagnetic PIC models

Tuesday, Nov 7, 2017 from 2PM to 3PM | POB 6.304

Important Update: NOTE: time change
  • Additional Information

    Hosted by Irene Gamba

    Sponsor: ICES Seminar

    Speaker: Eric Sonnendrücker

    Speaker Affiliation: Professor, Director of Numerical Methods in Plasma Physics Division, Max Planck Institute for Plasma Physics, Germany

  • Abstract

    A hamiltonian framework for the derivation of semi-discrete (continuous in time) Finite Element Particle In Cell approximations of the Vlasov-Maxwell equations was derived in [1]. It is based on a particle (Klimontovitch) discretization of the distribution function and a compatible Finite Element discretization of the grid quantities. The ideas introduced in [1]can be declined in di erent variants, choosing di erent discrete spaces for the elds or adding smoothing functions for the particles. Moreover, starting from such a semi-discretization, which yields a finite dimensional Hamiltonian structure de fined by a Poisson J (U) matrix and a hamiltonian H(U), several classes of di erent structure preserving time discretization can be derived: hamiltonian splitting methods as in [1], that preserve the Poisson structure, or discrete gradient methods that preserve exactly the hamiltonian. This procedure enables in particular to recover and generalize several well-known explicit and implicit PIC algorithms.

    We are going in this talk to give an overview of the geometric ideas behind this structure and how they can be used to derive fully discrete
    particle in cell schemes with exact conservation of the Poisson structure, the energy and Gauss' law.

    References
    [1] M. Kraus, K. Kormann, P.J. Morrison, E. Sonnendrucker. GEMPIC: Geometric electromagnetic particle-in-cell methods. Journal of Plasma Physics,83(4), (2017).


Tuesday, Nov 7

Challenges and Opportunities for Extracting Cardiovascular Risk Biomarkers

Tuesday, Nov 7, 2017 from 3:30PM to 5PM | POB 6.304

  • Additional Information

    Hosted by George Biros and Michael Sacks

    Sponsor: ICES Seminar - Computational Medicine Series

    Speaker: Ioannis Kakadiaris

    Speaker Affiliation: Professor, Computer Science, Electrical & Computer Engineering, and Biomedical Engineering, University of Houston

  • Abstract

    In this talk, I will present an analysis of the challenges and opportunities in the area of cardiovascular risk prediction. Specifically, I will present our research in the area of biomedical image computing with emphasis on the mining of information from cardiovascular imaging data for the detection of persons with a high likelihood of developing a heart attack in the near future (vulnerable patients). Emphasis will be given to methods for detection and segmentation of anatomical structures, and shape- and motion-estimation of dynamic organs. The left ventricle in non-invasive cardiac MRI (Magnetic Resonance Imaging) data is extracted using a new multi-class, multi-feature fuzzy connectedness method and deformable models will be/are used for shape and volume estimation. In non-invasive cardiac CT (Computed Tomography) data, the thoracic fat is detected using a relaxed version of the multi-class, multi-feature fuzzy connectedness method. Additionally, the calcified lesions in the coronary arteries are identified and quantified using a hierarchical supervised learning framework from the CT data. In non-invasive contrast-enhanced CT, the coronary arteries are detected using our tubular shape detection method for motion estimation and, possibly, for non-calcified lesion detection. In invasive IVUS imaging, our team has developed a unique IVUS acquisition protocol and novel signal/image analysis methods for the detection (for the first time in‐vivo) of ‘vasa vasorum’ (VV). The VV are micro-vessels that are commonly present to feed the walls of larger vessels. However, recent clinical evidence has uncovered their tendency to proliferate around areas of inflammation, including the inflammation associated with vulnerable plaques. In summary, our research is focused on developing innovative computational tools to mine quantitative parameters from imaging data for early detection of asymptomatic cardiovascular patients. The expected impact of our work stems from the fact that sudden heart attack remains the number one cause of death in the US, and unpredicted heart attacks account for the majority of the $280 billion burden of cardiovascular diseases.

    Bio
    Prof. Ioannis A. Kakadiaris is a Hugh Roy and Lillie Cranz Cullen University Professor of Computer Science, Electrical & Computer Engineering, and Biomedical Engineering at the University of Houston, Houston, TX, USA. He also holds an adjunct position at the School of Health Information Sciences at the University of Texas, Health Sciences Center. He joined UH in August 1997 after a postdoctoral fellowship at the University of Pennsylvania.

    Ioannis earned his B.Sc. in physics at the University of Athens in Greece, his M.Sc. in computer science from Northeastern University and his Ph. D. at the University of Pennsylvania. He is the founder and director of the Computational Biomedicine Lab. His research interests include cardiovascular informatics, biomedical image analysis, biometrics, computer vision, and pattern recognition.

    Dr. Kakadiaris is the recipient of a number of awards, including the NSF Early Career Development Award, Schlumberger Technical Foundation Award, UH Computer Science Research Excellence Award, UH Enron Teaching Excellence Award, and the James Muller Vulnerable Plaque Young Investigator Price. His research has been featured on Discovery Channel, National Public Radio, KPRC NBC News, KTRH ABC News, and KHOU CBS News

  • Multimedia

Friday, Nov 3

Recovery Guarantees for One-hidden-layer Neural Networks

Friday, Nov 3, 2017 from 10AM to 11AM | POB 6.304

  • Additional Information

    Hosted by Ivana Escobar and Sheroze Sheriffdeen

    Sponsor: ICES Seminar-Student Forum Series

    Speaker: Kai Zhong

    Speaker Affiliation: UT Austin

  • Abstract

    Neural Networks (NNs) have achieved great empirical success recently in various applications, including computer vision, natural language processing and reinforcement learning. However, due to the non-convexity of the NNs, the theoretical understanding of NNs is still limited. In this presentation, I will show that when inputs are sampled from Gaussian distribution and the activation function satisfies some properties, one-hidden-layer fully-connected NNs and one-hidden-layer convolutional NNs can be recovered in polynomial time. Specifically, I will first give a brief overview of recent theoretical progress on neural networks. Then we show gradient descent with tensor method initialization is guaranteed to converge to the ground truth parameters of the NNs with polynomial sample complexity and computational complexity.


Thursday, Nov 2

Posterior error control in Bayesian Inverse Problems

Thursday, Nov 2, 2017 from 10AM to 11:30AM | POB 6.304

Important Update: Please note TIME change
  • Additional Information

    Hosted by Tan Bui-Thanh

    Sponsor: ICES Seminar

    Speaker: J Andrés Christen

    Speaker Affiliation: Centro de Investigación en Matemáticas, Guanajuato, México

  • Abstract

    In the Bayesian analysis of Inverse Problems most relevant cases the forward maps are defined in terms of a system of (O, P)DE's that involve numerical solvers. These then lead to a numerical/approximate posterior distribution. Recently several results have been published on the regularity conditions required on such numerical methods to ensure converge of the numerical to the theoretical posterior. However, more practical guidelines are needed.

    I present some recent results that, by using Bayes Factors and in a finite dimensional setting, one can see that the numerical posterior tends to the theoretical posterior in the same order as the numerical solver used in the forward map. Moreover, when error estimates are available for the solver we can use a bound on this errors, proven to lead to basically error free posteriors. That is, given that we are observing noisy data, we may tolerate an amount (relative to the data noise) of numerical error in the solver, and end up with a basically error free posterior. In this talk I will show these results, present some examples in ODEs and PDEs and comment on the generalizations to the infinite dimensional setting.

    Bio
    Dr J Andres Christen has a BSc in mathematics from Universidad Nacional Autonoma de Mexico (1989, UNAM, Mexico city) and a PhD in mathematics from the University of Nottingham (1994, Nottingham, UK). His expertise is in Bayesian Statistics. Dr Christen has been working on the field for more than 25 years in applied as well as theoretical aspects of the discipline. His areas of application include ecology, paleoecolgy, environmental change, bioinformatics, among others. Moreover, recently at CIMAT he has helped to create a group on the study of inverse problems using Bayesian statistics (Bayesian UQ). In particular, Dr Christen and the group at CIMAT are working on the analysis of complex biological systems, epidemiology, sound wave scattering and other applications in physics, as well as some theoretical aspects involved in the practice of Bayesian UQ. Dr Christen holds a tenure position at CIMAT (part of the national network of research centers of CONACyT) since 2003 and is a Investigador Nacional, Sistema Nacional de Investigadores level III. He is a member of the International Society for Bayesian Analysis (ISBA), participated in the Scientific Committee of ISBA 2009{2012 and was the chair of the Local Organizing Committee,
    2014 ISBA World Congress. (Personal home page www.cimat.mx/~jac/. See also uq.cimat.mx for more details on the UQ group at CIMAT.)


Thursday, Nov 2

Mathematics for Cryo-Electron Microscopy

Thursday, Nov 2, 2017 from 3:30PM to 5PM | POB 6.304

  • Additional Information

    Hosted by George Biros

    Sponsor: ICES Seminar

    Speaker: Amit Singer

    Speaker Affiliation: Professor, Princeton University

  • Abstract

    Single particle cryo-EM recently joined X-ray crystallography and nuclear magnetic resonance spectroscopy as a high-resolution structural method for biological macromolecules (the 2017 Nobel Prize in Chemistry). Furthermore, cryo-EM has the potential to analyze compositionally and conformationally heterogeneous mixtures and, consequently, can be used to determine the structures of complexes in different functional states. The 3D-structure and the possible structural variability need to be determined from many noisy two-dimensional tomographic projections, whose viewing directions and in-plane rotations are unknown. In this lecture, the speaker will give an overview of the computational challenges in cryo-EM analysis and how he and others are trying to face them, focusing on 3D ab-initio modelling and the heterogeneity problem of determining structural variability.

    Bio:
    Dr. Singer is a Professor in the Department of Mathematics at Princeton University. His current research is focused on developing algorithms for three-dimensional structuring of macromolecules using cryo-electron microscopy. His mathematical interests are linear and non-linear dimensionality reduction of high dimensional data, signal and image processing, spectral methods, convex optimization and semidefinite programming. He is also a Professor in the Program in Applied and Computational Mathematics (PACM) and the Center for Statistics and Machine Learning (CSML) at Princeton.


Tuesday, Oct 31

  • Additional Information

    Hosted by Leszek Demkowicz

    Sponsor: ICES Seminar

    Speaker: Stefan Sauter

    Speaker Affiliation: Professor-Dr., Institut für Mathematik, Universität Zürich

  • Abstract

    In our talk we consider the numerical solution of the electric Maxwell equation by hp finite element methods. We will derive error estimates which are explicit in the mesh size h, the local polynomial degree p of the finite elements, and the wave number k. The stability analysis requires the combination of frequency splittings with Helmholtz and Hodge decompositions and the derivation of k-explicit estimates for three types of dual problems. We will explain the proof of the main result: The Galerkin discretization is pollution free provided the resolution condition: p>= C1 log k and k h/p <= C2 is satisfied for some constants C1 and C2.

    Bio:
    Prof.Dr. Sauter, completed his Habilitation in Mathematics in 1997, and received his Doctorate in 1993. He has been a Full professor for “Angewandte Mathematik”, Universität Zürich since 1999. Previously he worked at the Universität Leipzig. His awards include:
    1981 First prize, Bundeswettbewerb Mathematik; 1988-1990 Stipend of the “Studienstiftung des Deutschen Volkes”; 1993-1994 Stipend of the German research foundation; 1996 Oberwolfach Prize in “Applied Mathematics”; and 2000 Reprint of the paper “Is the pollution effect of the FEM avoidable for the Helmholtz equation considering high wave numbers” (jointly with I. Babuška) in SIAM Reviews.