Oden Institute Past Seminars

In-person seminars are held Tuesdays and Thursdays in POB 6.304 from 3:30-5:00 pm, unless otherwise noted. Zoom seminars usually occur during the same time period. Speakers include scientists, researchers, visiting scholars, potential faculty, and Oden Institute/UT Faculty or staff. Everyone is welcome to attend. Refreshments are served at in-person seminars at 3:15 pm.

Monday, Apr 12

Patient-specific Modeling of Hemodynamics: A Vision Talk

Monday, Apr 12, 2021 from 4PM to 5PM | Zoom Meeting

  • Additional Information

    Hosted by Michael Sacks

    Sponsor: Oden Institute Zoom Seminar - Dell Medical School, Department of Surgery and Perioperative Care; Department of Internal Medicine, Division of Cardiology; and Biomedical Engineering

    Speaker: C. Alberto Figueroa

    Speaker Affiliation: Edward B. Diethrich M.D. Professor in Biomedical Engineering and Vascular Surgery, University of Michigan

  • Abstract

    In this vision talk, Dr. Figueroa will provide an overview of his research portfolio, which combines medical imaging, computational methods, and machine learning in the cardiovascular field. Applications range from cardiovascular disease research to surgical planning, medical device optimization, and non-invasive diagnostics.

    Dr. Figueroa has a long track record of working closely with a broad range of experts in vascular surgery, cardiac surgery and cardiology (adult and pediatric), radiology, and neurology, to develop and apply computational tools for cardiovascular research. In addition to an overview of his research portfolio, Dr. Figueroa will also discuss his vision for engaging medical trainees with the field of computational medicine across Dell Medical School, the Oden Institute, and the Biomedical Engineering Department. This includes his vision for development of training grants.

    Lastly, ideas for collaborating on the strategic growth of Dell Medical through teaching, community engagement, and philanthropy will also be discussed.

    Bio
    Alberto Figueroa received his PhD in Mechanical Engineering at Stanford University, where he developed computational methods fluid structure interaction simulation of hemodynamics. His first academic appointment was a King’s College London in the UK, where he was Senior Lecturer in the Division of Biomedical Engineering and Imaging Sciences. Dr. Figueroa is currently the Edward B. Diethrich M.D. Professor in Biomedical Engineering and Vascular Surgery at the University of Michigan. His laboratory is focused on three main areas: 1) developing tools for advanced modeling of blood flow. His group develops the modeling software CRIMSON (www.crimson.software); 2) studying the link between abnormal biomechanical stimuli and cardiovascular diseases such as hypertension and thrombosis; 3) simulation-based surgical planning to aid with the optimal planning of cardiovascular surgeries.


Friday, Apr 9

High-performance sampling of generic determinantal point processes

Friday, Apr 9, 2021 from 10AM to 11AM | Zoom Meeting

  • Additional Information

    Hosted by Anna Yesypenko

    Sponsor: Oden Institute Zoom Seminar - Babuška Forum series

    Speaker: Jack Poulson

    Speaker Affiliation: Founder, Hodge Star Scientific Computing

  • Abstract

    Determinantal Point Processes (DPPs) were introduced by Macchi as a model for repulsive (fermionic) particle distributions, but their recent popularization is largely due to their usefulness for encouraging diversity in the final stage of a recommender system. The standard sampling scheme for finite DPPs is a spectral decomposition followed by an equivalent of a randomly diagonally-pivoted Cholesky factorization of an orthogonal projection, which is only applicable to Hermitian kernels and has an expensive setup cost. Researchers have begun to connect DPP sampling to LDL' factorizations as a means of avoiding the initial spectral decomposition, but existing approaches have only outperformed the spectral decomposition approach in special circumstances, where the number of kept modes is a small percentage of the ground set size.
    We show that trivial modifications of LU and LDL' factorizations yield efficient direct sampling schemes for non-Hermitian and Hermitian DPP kernels, respectively. Further, it is experimentally shown that even dynamically-scheduled, shared-memory parallelizations of high-performance dense and sparse-direct factorizations can be trivially modified to yield DPP sampling schemes with essentially identical performance.
    The software developed as part of this research, Catamari, https://gitlab.com/hodge_star/catamari, is released under the Mozilla Public License v2.0. It contains header-only, C++14 plus OpenMP 4.0 implementations of dense and sparse-direct, Hermitian and non-Hermitian DPP samplers. Its extension to high-precision homogeneous self-dual embedding interior point methods within the package Conic, https://gitlab.com/hodge_star/conic, is also briefly discussed.

    Bio
    Dr. Jack Poulson defended his dissertation on fast, distributed-memory quasi-direct solvers for heterogeneous 3D time-harmonic wave equations in 2012. He spent a brief postdoc in Stanford's math department focusing on fast, distributed-memory algorithms for applying Egorov operators and inverting so-called strongly-admissible H-matrices before spending a few years as an Assistant Professor at Georgia Tech and then Stanford University focusing on fast computation of pseudospectra, lattice reduction techniques, and conic interior point methods. He then spent a few years as a research scientist at Google Research working at the intersection of large-scale recommendation systems and natural language processing.


Friday, Apr 9

Streaming Kernel PCA with applications to Kernel Analog Forecasting

Friday, Apr 9, 2021 from 2PM to 3PM | Zoom Meeting

  • Additional Information

    Hosted by Shane McQuarrie

    Sponsor: Oden Institute Virtual Seminar - CSEM Student Forum series

    Speaker: Amelia Henriksen

    Speaker Affiliation: Ph.D. Student, Oden Institute, UT Austin

  • Abstract

    Kernel Principal Component Analysis (KPCA) plays a critical role in many modern pipelines for analyzing physical systems. However, naively computing kernel matrices and their subsequent principal components poses drastic issues in terms of scalability. In this presentation, we will discuss the benefits of streaming kernel principal component analysis––a field of research that adds critical flexibility to one of the most common tools in optimization. We will specifically demonstrate the considerable improvements that can be achieved by applying streaming KPCA to kernel analog forecasting (KAF) for dynamical systems.

    Bio
    Amelia Henriksen is a PhD candidate under Dr. Rachel Ward in the Oden Institute for Computational Engineering and Sciences. She received a Master of Science degree in Computational Science, Engineering and Mathematics from the University of Texas in 2019 and a Bachelor of Science degree in Applied and Computational Mathematics from Brigham Young University in 2015. Her dissertation work has focused primarily on improving theory and applications for streaming dimensionality reduction. In 2019, Amelia was a finalist in UT Austin’s three-minute thesis competition. She currently lives in Austin with her husband and son, where she enjoys volunteering with her local community and eating the spiciest queso she can get her hands on.

    (The CSEM Student Forum is a seminar series given by current CSEM graduate students to their peers. The aim of the forum is to expose students to each other's research, encourage collaboration, and provide opportunities to practice presentation skills. First- and second-year CSEM students receive seminar credit for attending.)


Thursday, Apr 8

Personalized Medicine: to Twin or not to Twin

Thursday, Apr 8, 2021 from 3:30PM to 5PM | Zoom Meeting

  • Additional Information

    Hosted by Michael Sacks

    Sponsor: Oden Institute Zoom Seminar

    Speaker: Mark Palmer

    Speaker Affiliation: Distinguished Scientist, Core Technologies group, Corporate Strategy & Scientific Operations, Medtronic

  • Abstract

    The digital transformation in healthcare has given new energy to the pursuit of personalized medicine and in particular the potential for the “holy grail” of the human Digital Twin. However, many medical experts are skeptical of the ability to model the patient through the combination of mechanistic and data driven models. The rationale is that the processes required to populate either type of model require data across the entire complexity and hierarchy of the human physiome and that collecting such data would be harmful or perhaps even fatal to the patient. Consequently, the conversation is turning away from the Human Digital Twin towards cohort-based modeling approaches as the middle ground between evidence-based and personalized medicine. In this, presentation we will explore the goals of the Digital Twin in non-healthcare applications and the parallels to patient care. I will also make the case for why we should continue to pursue the Human Digital Twin, look at recent successes, and identify characteristics of imminent opportunities.

    Bio
    Mark Palmer, M.D., Ph.D. is a Distinguished Scientist in the Core Technologies group within Corporate Strategy & Scientific Operations at Medtronic. In his current role, Mark leads an internal team of advanced modeling and simulation consultants servicing the global enterprise, manages external collaborations, and leads the strategy and platform technologies for realistic human simulation. Mark also serves as a member of the Enterprise A. I. Working Group and leads the Enterprise Modeling and Simulation Working Group where he regularly reports to the R&D Council, Clinical Research Council, Executive Committee and the Board of Directors on the long-range R&T Strategy for Modeling & Simulation. Mark’s expertise includes fully coupled multi-scale finite element methods, large deformation tissue mechanics, and clinical image-based modeling techniques. In 2020, Mark was named a Medtronic Technical Fellow for his outstanding contributions to the company’s technical excellence. Mark is passionate about advancing human simulation, virtual patient technologies, and digital evidence standards for the medical device industry.


Tuesday, Apr 6

Advances in sampling methods for electromagnetic inverse scattering problems

Tuesday, Apr 6, 2021 from 3:30PM to 5PM | Zoom Meeting

  • Additional Information

    Hosted by Tan Bui-Thanh

    Sponsor: Oden Institute Virtual Seminar

    Speaker: Dinh-Liem Nguyen

    Speaker Affiliation: Assistant Professor, Department of Mathematics, Kansas State University

  • Abstract

    Broadly speaking, inverse scattering problems are the problems of determining information about an object (scatterer) from measurements of the field scattered from that object. These problems have applications to such diverse areas as nondestructive testing, radar, geophysical exploration, and medical imaging. Sampling methods, which were initiated by Colton and Kirsch in 1996, have become a major and efficient approach for solving inverse scattering problems. These methods are fast, non-iterative, simple to implement, and essentially do not require a priori information about the unknown scatterer. In this talk, we will discuss some of our recent results on new types of sampling methods for solving electromagnetic inverse scattering problems.

    Bio
    Dinh-Liem Nguyen is currently an Assistant Professor in the Department of Mathematics at Kansas State University. His research interests lie in applied inverse problems, scattering theory, and scientific computing. He obtained his Ph.D. in Applied Math from Ecole Polytechnique (France) in 2013. After that, he spent three years as a postdoctoral assistant professor at the University of Michigan and one year as a postdoctoral fellow at the University of North Carolina at Charlotte.


Friday, Apr 2

  • Additional Information

    Hosted by Anna Yesypenko

    Sponsor: Oden Institute Zoom Seminar - Babuška Forum series

    Speaker: Clayton G. Webster

    Speaker Affiliation: Senior Scientist, Oden Institute, UT Austin; Distinguished Research Fellow, Lirio AI Research; and Department of Mathematics, Virginia Tech and Auburn University.

  • Abstract

    In this presentation, we will present both convex and non-convex minimization techniques for approximating complex functions in high dimensions. Of particular interest is the parameterized PDE setting, where the target function is smooth, characterized by a rapidly decaying orthonormal expansion, whose most important terms are captured by a lower (or downward closed) set. By exploiting this fact, we develop a novel weighted minimization procedure with a precise choice of weights, and a modification of the iterative hard thresholding method, for imposing the downward closed preference. Moreover, the recovery of the corresponding best approximation using our methods is established through an improved bound for the restricted isometry property and a new theory for non convex optimization. We will also present theoretical results that reveal our new computational approaches possess a provably reduced sample complexity compared to existing compressed sensing, least squares, and interpolation techniques. Numerical examples are provided to support the theoretical results and demonstrate the computational efficiency of the new weighted minimization method.

    Bio
    Clayton Webster is a Senior Scientist in the Oden Institute for Computational Engineering and Sciences at The University of Texas and a Distinguished Research Fellow at Lirio AI Research. He is also jointly appointed in the Department of Mathematics at Virginia Tech and Auburn University. Before these appointments, he was a Distinguished Professor in the Department of Mathematics at The University of Tennessee and a Distinguished Scientist and Group Leader in the Computational and Applied Mathematics Group at Oak Ridge National Laboratory. Previously, Dr. Webster has served as the Director of Quantitative Trading at NextEra Energy Resources, Power Trading LLC., and the John von Neumann Fellow at Sandia National Laboratories. In addition, his worked has earned him numerous accolades, including the Department of Energy Career Award as well as being appointed as a Frontiers of Science Fellow, by the National Academy of Sciences. Dr. Webster’s research has been supported by a variety of organizations, including the: US Department of Energy, US Department of Defense, National Science Foundational, and several US corporations. Clayton currently serves as Editor-in-Chief or Numerical Methods for PDEs and several national and international conference organizing committees as well as numerous editorial boards. He received his Ph.D. under the supervision of Prof. Max Gunzburger, in Mathematics from Florida State University in 2007. He also earned a M.Sc. and B.Sc. from McMaster University in 2003 and 2001 respectively.


Thursday, Apr 1

Multiscale Fluid Dynamics and Moment Theories Derived from the Boltzmann Equation

Thursday, Apr 1, 2021 from 3:30PM to 5PM | Zoom Meeting

  • Additional Information

    Hosted by Irene Gamba

    Sponsor: Oden Institute Virtual Seminar

    Speaker: Michael Abdelmalik

    Speaker Affiliation: Postdoctoral Fellow, Oden Institute, UT Austin

  • Abstract

    In this talk, we discuss a multiscale description of fluid dynamics provided by kinetic theory, which describes a fluid by a density distribution of its molecules as a function of time, position and velocity. We consider the evolution of the density distribution given by the Boltzmann equation (BE), which governs the transport and interaction of the fluid molecules. Of particular relevance in our account of multiscale fluid dynamics is the derivation of the Euler and Navier-Stokes-Fourier closure relations as the leading and first-order terms in the so-called Chapman-Enskog expansion of the density distribution in BE [1]. Higher-order closures derived from the Chapman-Enskog expansion, such as the Burnett equations, are unstable and ill-posed [2]. This leaves a gap – a long-standing one in theoretical and computational mathematics [3] – in our understanding of closure relations that follow from the atomistic view of BE to the laws of motion of continua. To address that gap we split this talk into two parts:

    In the first part of the talk, we use variational multiscale (VMS) analysis to derive a hierarchy of compressible fluid-dynamic closures from BE that include the Euler, Navier-Stokes-Fourier, and a new alternative to the Burnett equations. We proceed to show that the derived hierarchy of models, including the alternative to the Burnett equations, inherit an entropy inequality that is satisfied by solutions of the Boltzmann equation, rendering all stable and well posed. In the second part of the talk, we propose finite-element methods that leverage the kinetic formulations of the fluid-dynamic models for entropy stability.

    We conclude the presentation with numerical results and a discussion of future research directions.

    This work is in collaboration with F. Baidoo, L. Caffarelli, I.M. Gamba and T.J.R. Hughes.

    [1] Saint-Raymond, L. (2014). A mathematical PDE perspective on the Chapman–Enskog expansion. Bulletin of the American Mathematical Society, 51(2), 247-275.
    [2] Bobylev, A. V. (2006). Instabilities in the Chapman-Enskog expansion and hyperbolic Burnett equations. Journal of statistical physics, 124(2), 371-399.
    [3] Gorban, A., & Karlin, I. (2014). Hilbert’s 6th problem: exact and approximate hydrodynamic manifolds for kinetic equations. Bulletin of the American Mathematical Society, 51(2), 187-246

    Bio
    Michael Abdelmalik is a Peter O'Donnell, Jr. Postdoctoral Fellow at the Oden Institute in the University of Texas at Austin, working with Drs. I.M. Gamba and T.J.R. Hughes. Michael received his PhD degree at the Department of Mechanical Engineering as well as two MSc degrees, one in Applied Mathematics and the other in Mechanical Engineering, from the Eindhoven University of Technology where he worked with Dr. H. van Brummelen.

    During his PhD, Michael developed state-of-the-art kinetic-theory-based models as well as finite-element methods for rarefied fluid dynamics. Michael’s current research work includes: continuum theory derivation from, and computational methods for, kinetic theory such as the Boltzmann equation for fluids and quasilinear diffusion for plasma; and image-based computational methods for subject-specific bio-continuum mechanics.


  • Additional Information

    Hosted by Shane McQuarrie

    Sponsor: Oden Institute Virtual Seminar - CSEM Student Forum series

    Speaker: 1) Meghana Palukuri and 2) Bassel Saleh

    Speaker Affiliation: 1) 4th-year Ph.D. student in the CSEM program, working with Dr. Edward Marcotte and 2) 3rd-year PhD student, working with Dr. Omar Ghattas

  • Abstract

    Speaker 1: Meghana Palukuri. Protein complexes in an organism are groups of proteins that interact with each other to perform a particular function. These can be computationally predicted from a protein-interaction network using community detection algorithms. We predict 146 protein complexes potentially linked to SARS-COV2 that can be analyzed further to yield possible drug target sites and reveal insights into the mechanism of viral infection. These are a part of 1028 human protein complexes predicted from hu.MAP, a protein-interaction network with ~7k proteins (nodes) and ~15k interactions (edges) with our algorithm, Super.Complex v3.0, in an order of minutes on TACC’s Stampede2. We also briefly discuss future ideas including a reinforcement learning method for community detection and methods to further improve community embeddings.

    Bio
    Meghana Palukuri is a fourth-year Ph.D. student in the CSEM program, working with Dr. Edward Marcotte. Meghana holds undergraduate and masters degrees in chemical engineering from the Indian Institute of Technology Madras and has industrial experience through internships with Amazon and Schlumberger.

    Speaker 2: Bassel Saleh. In this work we aim to solve the Bayesian inverse problem in gravitational wave analysis in a multifidelity framework. Recent developments in the theory of multifidelity modeling for many-query problems demonstrate the potential for accelerating the solution of such problems by combining evaluations of multiple models of varying accuracies and costs. We apply this methodology to the broad and diverse class of models used for simulating gravitational wave signals, focusing on multifidelity importance sampling methods that sample temperized low-fidelity posteriors in order to estimate statistics of an expensive high-fidelity posterior.

    Bio
    Bassel Saleh is a third-year PhD student in Dr. Omar Ghattas' research group. Bassel studied at UT Austin for his undergraduate degree in physics and computer science. He enjoys research that involves mathematically interesting models and conceptually challenging physics.

    (The CSEM Student Forum is a seminar series given by current CSEM graduate students to their peers. The aim of the forum is to expose students to each other's research, encourage collaboration, and provide opportunities to practice presentation skills. First- and second-year CSEM students receive seminar credit for attending.)


Thursday, Mar 25

  • Additional Information

    Hosted by Thomas Yankeelov

    Sponsor: Oden Institute Virtual Seminar

    Speaker: Livia Schiavinato Eberlin

    Speaker Affiliation: Assistant Professor, William H. Tonn Professorial Fellow, Departments of Chemistry, Oncology & Diagnostic Medicine, The University of Texas at Austin; Department of Surgery, Baylor College of Medicine

  • Abstract

    Mass spectrometry is recognized as a powerful analytical technology to acquire molecular profiles of biological samples comprised of thousands of molecular ions with unparalleled sensitivity and chemical specificity. In my laboratory, we have developed the MasSpec Pen technology as a handheld device integrated to a mass spectrometer that allows detection of rich molecular profiles directly from in vivo and ex vivo tissues on clinically relevant timescales (<15 seconds). The complex molecular data generated by the MasSpec Pen are used in conjunction with machine learning methods to build statistical models capable of distinguishing disease states with high accuracies (92-98%, depending on tissue type). In this presentation, I will describe my team's effort developing, translating, and testing the MasSpec Pen technology and classification models to diagnose tissues in the laboratory as well as intraoperatively, including challenges and opportunities to improve data analysis and statistical classification.

    Co-hosted by Tom Yankeelov and Jack Virostko.

    Bio
    Prof. Livia Schiavinato Eberlin is an Assistant Professor in the Departments of Chemistry, Oncology and Diagnostic Medicine at the University of Texas at Austin, and an adjunct Assistant Professor in the Department of Surgery at Baylor College of Medicine. Dr. Eberlin received her B.S. in Chemistry in 2008 from the State University of Campinas, her Ph.D. in Analytical Chemistry in 2012 from Purdue University under the guidance of Prof. Graham Cooks, and pursued her postdoctoral research with Prof. Richard Zare in the Department of Chemistry at Stanford University. In 2016, Prof. Eberlin started her independent career at The University of Texas at Austin. Her research in mass spectrometry has been recognized through grants and awards, including a NIH K99/R00 Pathway to Independence Award, a Moore Inventor Fellowship, and a MacArthur Fellowship. Her research program centers around the development and application of novel mass spectrometry technologies in health-related research, with a particular focus on disease detection and diagnosis.


  • Additional Information

    Hosted by Greg Rodin

    Sponsor: Oden Institute Virtual Seminar

    Speaker: Joseph Bishop

    Speaker Affiliation: Principal Member of Technical Staff, Sandia National Laboratories

  • Abstract

    Modeling the macroscale response of a structure requires the use of a material constitutive model that provides the effective or homogenized behavior of the underlying material microstructure. Assuming that homogenized properties exist for the given microstructure, there is also an inherent assumption of a separation of scales when the properties are used to predict the macroscale response of a structure. There are several engineering applications in which these assumptions may be violated, in particular for metallic structures obtained through additive manufacturing. Instead of resorting to direct numerical simulation of the macroscale system with an embedded fine scale, an alternative approach is to use an approximate macroscale constitutive model, but then estimate the model-form error using a posteriori estimation techniques and subsequently adapt the macroscale model to reduce the error for a given boundary value problem and quantity of interest. We investigate this approach to multiscale analysis in solids with unseparated scales using the example of an additively-manufactured metallic structure consisting of a polycrystalline microstructure that is neither periodic nor statistically homogeneous. As a first step to the general nonlinear case, we focus here on linear elasticity in which each grain within the polycrystal is linear elastic but anisotropic.

    Bio
    Joe Bishop received his Ph.D. in Aerospace Engineering from Texas A&M University in 1996. His graduate research was in the mechanics of composite materials and mechanisms of material damping. From 1997 to 2004 he worked in the Synthesis & Analysis Department of the Powertrain Division of General Motors Corporation, performing thermal-structural analysis of internal combustion engines with a focus on predicting high-cycle fatigue performance of the base engine. He joined Sandia National Laboratories in 2004 in the Engineering Sciences Center. He has worked on diverse topics including pervasive-fracture, impact and penetration, geologic CO2 sequestration, metal additive manufacturing, residual stress measurement techniques, polyhedral finite element formulations, meshfree methods, and multiscale simulations in solid mechanics. He is currently the manager of a modeling and simulation department within the Engineering Sciences Center at Sandia.