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.

Friday, Oct 12

Explainable AI: Making Visual Question Answering Systems more Transparent

Friday, Oct 12, 2018 from 10AM to 11AM | POB 6.304

  • Additional Information

    Hosted by Tom O'Leary-Roseberry and Kendrick Shepherd

    Sponsor: ICES Seminar-Babuska Forum Series

    Speaker: Raymond J. Mooney

    Speaker Affiliation: Department of Computer Science, UT Austin

  • Abstract

    Artificial Intelligence systems’ ability to explain their conclusions is crucial to their utility and trustworthiness. Deep neural networks have enabled significant progress on many challenging problems such as visual question answering (VQA), the task of answering natural language questions about images. However, most of them are opaque black boxes with limited explanatory capability. The goal of Explainable AI is to increase the transparency of complex AI systems such as deep networks. We have developed a novel approach to XAI and used it to build a high-performing VQA system that can elucidate its answers with integrated textual and visual explanations that faithfully reflect important aspects of its underlying reasoning while capturing the style of comprehensible human explanations. Crowd-sourced human evaluation of these explanations demonstrate the advantages of our approach.

    Bio:
    Raymond J. Mooney is a Professor in the Department of Computer Science at the University of Texas at Austin. He received his Ph.D. in 1988 from the University of Illinois at Urbana/Champaign. He is an author of over 170 published research papers, primarily in the areas of machine learning and natural language processing. He was the President of the International Machine Learning Society from 2008-2011, program co-chair for AAAI 2006, general chair for HLT-EMNLP 2005, and co-chair for ICML 1990. He is a Fellow of the American Association for Artificial Intelligence, the Association for Computing Machinery, and the Association for Computational Linguistics and the recipient of best paper awards from AAAI-96, KDD-04, ICML-05 and ACL-07.


Thursday, Oct 11

Collaborative Opportunities with Applied Research Laboratories

Thursday, Oct 11, 2018 from 3:30PM to 5PM | POB 6.304

  • Additional Information

    Hosted by Karen Willcox

    Sponsor: ICES Seminar

    Speaker: Marcia J. Isakson

    Speaker Affiliation: IR&D Strategic Program Director, Research Scientist, Applied Research Laboratories, The University of Texas at Austin

  • Abstract

    The Applied Research Laboratories (ARL:UT), located on the Pickle Research Campus, is a University Affiliated Research Center (UARC) originally organized in 1945. The laboratories have over 400 researchers with over $100M per year in funding primarily in the areas of acoustics, electromagnetics, and information technology. Although greater than 90% of the projects are from the department of defense or the intelligence community, there are projects from the oil and gas industry and the medical community as well. This presentation will offer an overview of ARL:UT and highlight projects and areas that would benefit from collaboration with ICES in the areas of reduced order modeling, finite element modeling, inversion, control systems, autonomy and machine learning. The projects include environmental inversion using acoustic propagation measurements, aeroacoustics for jet engine and propeller blade noise, autonomy algorithms for unmanned vehicles, finite element modeling for acoustic interaction with ocean sediments and an overview of the machine learning community at ARL. After the presentation, project leaders will be available to meet with ICES faculty.

    Bio
    Dr. Isakson received her B.S. in engineering physics and mathematics from the United States Military Academy at West Point in 1992. Upon graduation, she was awarded a Hertz Foundation Fellowship and completed a master’s degree in physics from the University of Texas at Austin in 1994. CPT Isakson served in the United States Army from 1994-1997 as a battalion operations officer. She earned a Ph.D. in Physics from the University of Texas at Austin in 2002. She has been employed at the Applied Research Laboratories at the University of Texas (ARL:UT) as a research scientist since 2001. In 2018, Dr. Isakson became the IR&D and Student Program Coordinator and Strategic Research Director for ARL:UT. Dr. Isakson is a Fellow of the Acoustical Society of America (ASA) and a Member and Distinguished Lecturer of the IEEE, Oceanic Engineering Society. Dr. Isakson currently serves as the past-president of the ASA and on the governing board of the American Institute of Physics (AIP).

    Dr. Marcia J. Isakson, Ph.D.
    IR&D Strategic Program Director
    Applied Research Laboratories
    The University of Texas at Austin
    misakson@arlut.utexas.edu
    (512) 835-3790


Thursday, Oct 11

Artistic Practices and Scientific Research: Interdisciplinary Panel Discussion

Thursday, Oct 11, 2018 from 6PM to 9PM | Art Building, Rm. 1.102

Important Update: PLEASE Note different venue and times.
  • Additional Information

    Hosted by Patrick Heimbach and An T. Nguyen

    Sponsor: Joint ICES/Department of Art and Art History (DAAH)

    Speaker: Panelists listed below.

    Speaker Affiliation: Panelists affiliations are listed below.

  • Abstract

    Join us for an interdisciplinary panel discussion with artists Annesofie Norn and William Trossell (ScanLAB Projects), curator Markus Reymann (TBA 21-Academy), and oceanographer Patrick Heimbach (The University of Texas at Austin), moderated by Ulrike Heine, curator of Exploring the Arctic Ocean. A reception will follow.

    Organized by the Institute for Computational Engineering and Sciences (ICES) and the Department of Art and Art History, The University of Texas at Austin. Presented in conjunction with the exhibition Exploring the Arctic Ocean (http://sites.utexas.edu/utvac/exploring-the-arctic-ocean/).

    Panelists:
    •Annesofie Norn is a visual artist, scenographer and curator working at the intersection of art, humanitarian development, and public engagement. Her works, which often involve situated engagement, investigate modes of mapping, measuring, and observation. Norn currently leads a long-term project for the United Nations, in which she explores the potential of participative storytelling together with internally displaced families in Gaza. Together with Ole Kristensen and Daniel Plewe, Norn produced the documentary and video installation “Longing Fast Forward,” which is on view in Exploring the Arctic Ocean.

    •William Trossell is an artist, architect, and technologist. He is the cofounder of ScanLAB Projects, a pioneering creative practice that is half art studio and half research laboratory. From 2011 to 2013, while working for the Climate Impact Tour with Greenpeace and Cambridge University aboard the icebreaker The Arctic Sunrise, ScanLAB Projects captured 26 ice floes in forensic detail in the Fram Strait using LIDAR 3D scanning technology. Replicas of the floes and other forensic records were exhibited in Frozen Relic: Arctic Works at the AA Gallery, Royal Academy and the Louisiana Museum of Modern Art. Parts of the project are on view in Exploring the Arctic Ocean.

    •Markus Reymann is the director and cofounder of TBA21-Academy. Since July 2011, he has initiated and conducted numerous expeditions; each trip is designed as a collaboration with invited artists, scientists, and thinkers who are eager to embark on oceanic explorations. In December 2015 at the 21st Conference of Parties (COP21) in Paris, Reymann announced TBA21-Academy’s latest program, “The Current,” which was conceived to raise awareness of today’s most urgent ecological, social, and economic issues.

    •Patrick Heimbach is an associate professor at The University of Texas at Austin with appointments in the Institute for Computational Engineering and Sciences, the Institute for Geophysics, and the Jackson School of Geosciences. His research focuses on the ocean circulation and its role in climate, as well as on the dynamics of the polar ice sheets, oceans, and sea ice. He directs the Computational Research in Ice and Ocean Systems (CRIOS) group. Heimbach is a member of numerous international research consortia and has been an invited scientist to convenings of TBA21-Academy.

    •Ulrike Heine is a visual studies scholar and curator with a focus on the intersection of visual arts and ecology. She completed her PhD with a thesis on climate change-related imagery. From 2013 to 2016, Heine was on the curatorial staff of the MIT Museum in Cambridge, an institution working at the intersection of science, technology, and the arts. She curated "Exploring the Arctic Ocean" (http://sites.utexas.edu/utvac/exploring-the-arctic-ocean/" in collaboration with An T. Nguyen (ICES) and Patrick Heimbach.

    Note: The link input in the Media Section below does not stream.

  • Multimedia

     Event Stream Link: Click Here to Watch


  • Additional Information

    Hosted by Ufuk Topcu

    Sponsor: ICES Seminar

    Speaker: Matthew Clark

    Speaker Affiliation: Galois, Inc.

  • Abstract

    With the continued rise in the interest of Machine Learning and AI technologies for DoD applications, this talk will briefly cover my perspective on the terms Autonomy, AI, Machine Learning as they relate to future DoD opportunities. The perspective will start from the "Three Waves of AI" discussion by John Launchbury while he was the DARPA I2O director. From that perspective, this talk will briefly cover the up and coming challenges, research opportunities and some particular agencies and industry partners that plan on increasing their investment in AI significantly over the next 5 years.

    Bio
    At Galois, Mr. Clark focuses on the design and assurance of intelligent cyber-physical systems. His research interests lie in first-principles approaches to the development of hybrid systems, artificial intelligence, fuzzy logic, run-time assurance protected systems, robotics and control. Before joining Galois, Mr. Clark was previously the lead for Autonomous Manned-Unmanned Teaming for the Aerospace Systems Directorate, Air Force Research Laboratory (AFRL). During his tenure at AFRL, Mr. Clark also served as the supervisor of the Autonomous Controls Branch, Power and Control Division, Aerospace Systems Directorate, where he received the Air Force Meritorious Civilian Service Award. Prior to that he was the Technical Area Lead for the verification and validation of complex and autonomous systems. He led a team of in-house researchers in the design, analysis, verification, and validation of autonomous control systems. He was also considered the primary subject matter expert for AFRL Test and Evaluation, Verification and Validation (TEV&V) of complex systems and the DoD co-leader and primary author of the Test and Evaluation Strategy for the Assistant to the Secretary of Defense, Research and Evaluation, Autonomy Community of Interest, (ASD/R&E COI).


Tuesday, Oct 9

Generalized Source Integral Equations

Tuesday, Oct 9, 2018 from 3:30PM to 5PM | POB 6.304

  • Additional Information

    Hosted by Ali Yilmaz

    Sponsor: ICES Seminar

    Speaker: Amir Boag

    Speaker Affiliation: Professor, Department of Physical Electronics, School of Electrical Engineering, Faculty of Engineering, Tel-Aviv University

  • Abstract

    The Generalized Source Integral Equations (GSIEs) is a family of integral formulations which are designed to be inherently compressible, for the purpose of developing general fast direct solvers for arbitrary shaped essentially convex scatterers. These formulations employ highly directional sources with complex radiation patterns, rather than the conventional non-directional sources. Two types of directive sources designed to achieve deep shadow in a prescribed angular sector covering the interior of a scatterer have been developed. First type uses electric and magnetic currents to absorb the radiation of an isotropic elemental source within the specified angular sector, while the second - employs perfectly conducting elliptical shields to deflect the undesirable radiation. Such sources inherently eliminate line-of-sight interactions between the opposite sides of an essentially convex scatterer, thus effectively reducing the problem’s dimensionality. The dimensionality reduction leads in turn to multilevel compressible matrices and facilitates the construction of fast direct solvers.

    Bio
    Amir Boag received the B.Sc. degree in electrical engineering and the B.A. degree in physics in 1983, both Summa Cum Laude, the M.Sc. degree in electrical engineering in 1985, and the Ph.D. degree in electrical engineering in 1991, all from Technion - Israel Institute of Technology, Haifa, Israel. From 1991 to 1992 he was on the Faculty of the Department of Electrical Engineering at the Technion. From 1992 to 1994 he has been a Visiting Assistant Professor with the Electromagnetic Communication Laboratory of the Department of Electrical and Computer Engineering at the University of Illinois at Urbana-Champaign. In 1994, he joined Israel Aircraft Industries as a research engineer and became a manager of the Electromagnetics Department in 1997. Since 1999, he is with the Physical Electronics Department of the School of Electrical Engineering at Tel Aviv University, where he is currently a Professor. Dr. Boag's interests are in computational electromagnetics, wave scattering, imaging, and design of antennas and optical devices. He has published over 110 journal articles and presented more than 250 conference papers on electromagnetics and acoustics. Prof. Boag is an Associate Editor for IEEE Transactions on Antennas and Propagation. He is a Fellow of the Electromagnetics Academy. In 2008, Amir Boag was named a Fellow of the IEEE for his contributions to integral equation based analysis, design, and imaging techniques.


Tuesday, Oct 2

Low-rank approximation and rank-revealing for oscillatory-kernel matrix blocks

Tuesday, Oct 2, 2018 from 3:30PM to 5PM | POB 6.304

  • Additional Information

    Hosted by Ali Yilmaz

    Sponsor: ICES Seminar

    Speaker: Yaniv Brick

    Speaker Affiliation: Senior Lecturer (Assistant Prof.), Department of Electrical and Computer Engineering, Ben-Gurion University of the Negev

  • Abstract

    Various fast direct integral equation solvers rely on the low-rank (LR) representation of sub-matrices (blocks) of the boundary elements method (BEM) matrix or moment matrix Z. In particular, direct solvers based on the hierarchical-matrix approach use LR approximate factorization Z_os≈AB^T of “admissible” blocks Z_os corresponding to interactions between separate groups (clusters) of basis (source) and testing (observer) functions that are sufficiently separated. The straightforward algebraic LR factorization, e.g., by using the singular value decomposition (SVD), involves O(N^2) time and memory for generating and storing the blocks, and an O(N^3) time for applying the algebraic procedure itself, and becomes a computational bottleneck for any such solver. While for non-oscillatory kernels many alternative algorithms, including some kernel independent alternatives, enable the fast LR factorization of MoM matrix blocks at an O(N) cost, accurate revealing of the rank and computing the LR factorization beyond the quasi-static regime is performed more efficiently by using physics-based methods.

    In my talk, I will discuss a class of hierarchical algorithms, developed together with the Computational Electromagnetics Group at UT, for fast LR approximation of oscillatory kernel moment matrix blocks. These rely on the multilevel partitioning of the source cluster associated with Z_os into a tree of sub-clusters. This enables the application of the SVD not directly to Z_os but to smaller matrices, which represent interactions between the source sub-clusters and observers on auxiliary grids; phase- and amplitude-compensation of the interactions allows for coarse (non-uniform) sampling from which the block Z_os can be reconstructed approximately, at the hierarchy’s top level. We will present two variants of the algorithm: (i) a fast algorithm for the computation of Z_os≈AB^T that makes use of volumetric non-uniform grids, and has been shown to effectively achieve computation times that are ∝N^3/2 and ∝N^2 for very large quasi-planar and densely packed problems (with an expected asymptotic costs of O(N^2) and O(N^7/3), respectively) and storage of O(N^3/2). (ii) an even faster algorithm for the computation of an orthonormal column matrix B and the rank only, at costs of N^3/2 and N^2 (for quasi-planar and densely packed cases, respectively), followed by the computation of Z_osB≈A that can be accelerated by employing fast matrix-vector multiplication methods. The performance of the two variants, in terms of memory, computation time, and rank-revealing accuracy, will be demonstrated and compared via representative examples, and their relevance for various applications/problems will be discussed.

    Bio
    Yaniv Brick received the B.Sc. (magna cum laude), M.Sc. (summa cum laude), and Ph.D. degrees in electrical engineering from the School of Electrical Engineering, Tel Aviv University, Tel Aviv, Israel, in 2005, 2007, and 2015, respectively. From 2014 to 2017, he was a Post-Doctoral Fellow at the Institute for Computational Engineering and Sciences, The University of Texas at Austin. Since 2017, he is with the Department of Electrical and Computer Engineering, Ben-Gurion University of the Negev, Beersheba, Israel. His current research interests include wave theory and numerical modeling for electromagnetic and acoustic analysis, with an emphasis on fast algorithms for direct and iterative solution of integral equations. Dr. Brick was a recipient of the IEEE Antennas and Propagation SocietyDoctoral Research Award in 2012, the Peter O’Donnell, Jr. Postdoctoral Fellowships in Computational Engineering and Sciences in 2014–2016, the Fulbright Postdoctoral Fellowship Program grant in 2014–2015, and the 2015 Fulbright Alumni Prize.


  • Additional Information

    Hosted by Tom O'Leary-Roseberry and Kendrick Shepherd

    Sponsor: ICES Seminar-Babuska Forum Series

    Speaker: Karen Willcox

    Speaker Affiliation: Director, ICES, UT Austin

  • Abstract

    Analysis of the physical governing equations of a system can reveal variable transformations that transform a general nonlinear model into a model with more structure. In particular, the introduction of auxiliary variables can convert a general nonlinear model to a model with polynomial nonlinearities, a so-called "lifted" system. The lifted model is equivalent to the original model; it uses a change of variables, but introduces no approximations. We present an approach that combines lifting with proper orthogonal decomposition model reduction. The approach uses a data-driven formulation to learn the low-dimensional model from high-fidelity simulation data, but a key aspect of the approach is that the state-space in which the learning is achieved is derived using the problem physics. A key benefit of the approach is that there is no need for additional approximations of the nonlinear terms, in contrast with existing nonlinear model reduction methods that require sparse sampling or hyper-reduction. A second benefit is that the lifted problem structure opens new pathways for rigorous analysis and input-independent model reduction. The method is demonstrated for nonlinear systems of partial differential equations arising in rocket combustion applications.
    ** Note that Professor Willcox has an open GRA position for research related to this topic.

    Bio
    Karen E. Willcox is Director of the Institute for Computational Engineering and Sciences (ICES) and a Professor of Aerospace Engineering and Engineering Mechanics at the University of Texas at Austin. She holds the W. A. “Tex” Moncrief, Jr. Chair in Simulation-Based Engineering and Sciences and the Peter O'Donnell, Jr. Centennial Chair in Computing Systems. Prior to joining ICES in 2018, she spent 17 years as a professor at the Massachusetts Institute of Technology, where she served as the founding Co-Director of the MIT Center for Computational Engineering and the Associate Head of the MIT Department of Aeronautics and Astronautics. Prior to joining the MIT faculty, she worked at Boeing Phantom Works with the Blended-Wing-Body aircraft design group. Her research at MIT has produced scalable computational methods for design of next-generation engineered systems, with a particular focus on model reduction as a way to learn principled approximations from data and on multi-fidelity formulations to leverage multiple sources of uncertain information. She is a Fellow of SIAM and Associate Fellow of AIAA.


Friday, Sep 28

  • Additional Information

    Hosted by Takashi Tanaka

    Sponsor: ICES Seminar

    Speaker: Yorie Nakahira

    Speaker Affiliation: California Institute of Technology

  • Abstract

    The neurosciences provide rich, diverse details on how humans sense/communicate/compute/actuate movement using efficient, distributed hardware with tradeoffs in sparsity, quantization, noise, delays, and saturation throughout. These processes are implemented in highly-layered architectures involving high-level goals/plans/decisions and low-level sensing/reflex/action to facilitate robust control. Missing is an integrative view that connects component-level tradeoffs/constraints with sensorimotor performance and effective architectures. In this talk, we briefly review essential neuroscience motivation, emphasizing speed/accuracy tradeoffs (SATs). SATs are among the most extensively studied and ubiquitous tradeoffs in both neurophysiology and sensorimotor control literature. We model the component SATs in spiking neuron communication and their sensory and muscle endpoints. We then provide both stochastic and deterministic frameworks that yield tight analytic bounds on how component SATs impose sensorimotor control SATs. From the resulting optimal control policies, we clarify the benefit of layering and heterogeneities in neurons, muscles, and sensorimotor control loops. We also briefly sketch our new experimental platforms and experiments that illustrate the theory and highlight tradeoffs and layering. Finally, we show that the optimal controller structures match the cryptic patterns of feedback and feedforward seen in vertebrate nervous systems.

    Bio
    Yorie Nakahira is a Ph.D. student at California Institute of Technology. Her primary research interests are control and information theory with applications to neuroscience and biology


Thursday, Sep 27

  • Additional Information

    Hosted by Tan Bui-Thanh

    Sponsor: ICES Seminar

    Speaker: Roland Glowinski

    Speaker Affiliation: Cullen Professor of Mathematics and Mechanical Engineering, University of Houston

  • Abstract

    Monge-Ampère equations occur areas of Science and Engineering: Differential Geometry, Fluid Mechanics, Elasticity, Cosmology, Antenna and Windshield Design, Mesh Generation, Optimal Transport, Finance, Image Processing, etc.

    Our goal in this presentation is discuss the numerical solution of the canonical Monge-Ampère equation in dimensions 2 and 3. These last two-decades, the numerical solution of Monge-Ampere equation (MA-D) has motivated a relatively large number of publications, however most of the solution methods we know use either high order finite element approximations or wide-stencil finite difference ones. The method we going to present relies on: (i) piecewise affine approximations of the unknown function u and of its second order derivatives, (ii) a time discretization by operator-splitting of an initial value problem associated with an equivalent divergence formulation of problem (MA-D), (iii) a projection operator on the cone of the symmetric positive semi-definite matrices, (iv) a Tychonoff regularization method to approximate the second order derivatives. The resulting methodology is modular and can handle easily domains with curved boundaries. It is also robust in the sense that it can also handle efficiently non- smooth situations, the non-smoothness coming from the data (if, for example, forcing is a the positive multiple of Dirac measure), or from data incompatibility. The results of numerical experiments will be presented, including those associated with the solution of the following (nonlinear) eigenvalue problem for the Monge-Ampère operator.


  • Additional Information

    Hosted by Ron Elber

    Sponsor: ICES Seminar-Molecular Biophysics Series

    Speaker: Thomas Bishop

    Speaker Affiliation: Louisiana Tech University

  • Abstract

    Historically, bioinformatics and computational biology are recognized as distinct endeavors. The underlying theories, experiments, software and computing resources differ significantly. We demonstrate that these differences can be overcome by exploiting existing data standards, algorithms, and web based tools to study the structure of DNA, nucleosomes and chromatin in atomic and coarse grain detail from single base pairs to megabase regions of chromatin and beyond.
    In this presentation I will define the “genomics dashboard” concept, explain the mathematics and algorithms that drive G-Dash , our prototype genome dashboard, then demonstrate how experimentally determined maps of nucleosome positions for Saccharomyces cerevisiae can be used to assemble a computational karyotype. Comparative all atom molecular dynamics simulations of nucleosomes and coarse-grained models of the MMTV, CHA1, HIS3 and PHO5 promoters highlight important observations. DNA kinking in the nucleosome depends on both sequence and position. Experimentally determined nucleosome positions are insufficient to achieve tight packing of chromatin. Sequence specific material properties of DNA (conformation & flexibility) can affect chromatin bending and looping.