Dr. Silvia Crivelli is a scientist at Berkeley National Lab. She received her Ph.D. in Computer Science from the University of Colorado, Boulder in 1995. She was a postdoctoral fellow at the University of California, Berkeley, and the Lawrence Berkeley National Laboratory (LBNL), where she began to work in computational biology. Currently, she is a Program Manager 2 at LBNL and an Associate Researcher at UC Davis. Her research interests are two-fold: to bring scientists together, both seasoned and young and from all walks of science, to tackle long-standing, extremely hard, and multidisciplinary biological problems and to develop highly interactive information modeling tools that empower physicians and researchers to explore and understand the mechanisms of health and disease.
Dr. William F Godoy joined ORNL in 2016, currently a Senior Computer Scientist in the Computational Science and Mathematics Division, CSMD. A native of Peru, his background is in Mechanical Engineering with a strong research and development (R&D) focus on the computational aspects of large-scale scientific modeling and simulation. Dr. Godoy obtained his Ph.D. (2009) and MS (2006) degrees from the State University of New York (SUNY) at Buffalo, working under the supervision of Dr. Paul E. DesJardin on 3D radiative transfer models for fire suppression systems. Upon graduation, he joined NASA Langley Research Center for a postdoctoral appointment (2009-2012) to study radiative transfer in the atmosphere using NASA's Pleiades supercomputer and GPU computing. Prior to ORNL, he worked as a Senior Software Engineer at Intel Corporation (2012-2016) on the R&D of large-scale thermal reliability models to support semiconductor design and fabrication of Intel's 22 nm, 14 nm (Haswell, Skylake, and Knights Landing) technologies, and their 10 nm and 7 nm successors.
Dr. Linda Allen will share her backstory and will have an open Q&A with students.
Lunch and soft drinks will be served.
Computational Models of Self-Propelled Microorganisms
Abstract
Many interesting biological phenomena involve rigid or flexible thin filaments interacting with a fluid. Some examples are the motion of bacteria swimming through the actuation of flagella, the coordinated motion of cilia, the swimming of spermatozoa, and artificial self-propelled microswimmers. Of particular interest is the interaction of filaments with nearby surfaces or moving through regions in the fluid containing elastic polymers, since these environments commonly found in nature. I will present work on computational models of microscopic filaments moving in a fluid based on recent advances of the method of regularized Stokeslets. The method is based on fundamental solutions of linear partial differential equations, modified to remove the singularities. I will show results from simulations of flagellar motions with asymmetric beat patterns and the effect of swimming near a solid surface. Recent work on models of swimming through viscoelastic regions incorporate the effect of the elastic polymers immersed in the fluid using a network of cross-linked nodes where each link is modeled by a simple viscoelastic element. The presentation will be expository; no previous experience in computational fluid dynamics or biological flows is necessary.
Computer vision aims to give machines the ability to see. If we want industrial pick-and-place robots, drill robots in enabled free-areas, self-driving cars, and lifelike assistants that perform natural speech patterns, we must build machines with the same visual capabilities that humans enjoy. So, how can we teach machines to see?
Computer vision systems examine each and every pixel in an image to determine whether or not a specific feature is present, this is known as feature extraction. Model-based methods and data-driven methods are the two broad categories of feature extraction methods. Model-driven approaches involve hand-coding features one at a time, whereas, in data-driven approaches, the most descriptive features of each object definition are determined by the algorithms/models themselves.
Whether or not a computer vision problem is best solved with model-driven or data-driven approaches depends on several factors, including the access to data, hardware or device resources, the full range of data cases and expected variations. Therefore, the goal of this talk is deepen data cases, the design and adaptation of data-driven approaches for automating clinical processes.
Come join us to listen to Dr. Juan Mayorga-Zambrano, who is a titular professor of Mathematics at Yachay Tech University. This talk will include his last works on compact embeddings. In particular, the p-Sobolev-like cone of operators is compactly embedded in the self-adjoint trace-class operators on L2. In the path, some regularity properties for the density function associated with self-adjoint trace-class operators on L2 as well as Gagliardo-Nirenberg type inequalities departing from Lieb-Thirring type conditions are proved. Moreover, the compactness property is applied to minimize free energy functionals where the entropy term is generated by a Cassimir-class function related to the eigenvalue problem of the Schrödinger operator −α∆ + V, α > 0, with Dirichlet condition. You will be able to discuss and ask questions to Dr. Mayorga-Zambrano about his current work at Yachay Tech University and his career path. This talk will offer further knowledge enrichment to students in STEM.
Come join us to listen to Sam Jacobs, who is a computer scientist at Lawrence Livermore National Laboratory. This talk will Progress in deep learning is anchored on a tripod: increased computational power, more sophisticated neural network architectures, and growing datasets. Tremendous advances have been made, but we are yet to maximize what is possible. Training of neural networks for vision and language model applications is inhibited by the prohibitive cost of model exploration and model training. Scientific applications of interest to the US Department of Energy and US national laboratories face additional challenges: (1) scientific data are often hard to label and require domain expertise, (2) the measure of uncertainty required by predictive models, and (3) satisfying scientific constraints. Given these requirements and others, neural network architectures must be frequently tuned for specific scientific applications requiring extensive architecture search and hyperparameter tuning. In this talk, I will present our ongoing work in large-scale training and architecture search of deep neural networks. I will discuss different parallelization techniques in an HPC-centric deep learning toolkit- Livermore Big Artificial Neural Networks (LBANN)- and its extension for fast, robust, and scalable neural architecture search. Experimental results from different application domains including small-molecule antiviral drug design for CoVID-19 will be presented. You will be able to discuss and ask questions to Dr. Jacobson about his current work at Yachay Tech University and his career path. This talk will offer further knowledge enrichment to students in STEM.
Abstract: Any physical theory relies on a mathematical model, typically in the form of a Partial Differential Equation (PDE), the solutions of which yield the predictions of the theory. Barring some simple cases, the solutions are found numerically by devising efficient algorithms. However, this task becomes harder as the physical domain becomes more complicated. Conventional techniques typically yield low order accuracy and end up requiring a balance between accuracy, and computational effort. In this talk, I will discuss some of my efforts towards developing numerical methods for complex geometries. One such common yet largely unexplored domain is the three-dimensional cylinder. Employing carefully chosen basis functions, leads to powerful solvers and allows us to address challenges such as the nonlinear Faraday wave problem. Next, we shall move beyond domain-specific approaches and identify the subtle issues underlying the shortcomings of the generic low accuracy methods. After providing an overview of some of the proposed methods to overcome these issues, I shall present the Projection Extension method, a novel approach that yields highly accurate solutions on a large array of complex domains. The technique leverages simple ideas to great effect, with the result that it is straightforward to apply, implement, and generalize, and allows us to compute demonstrably accurate solutions to several challenging models, including heat flow and Newtonian and viscoelastic fluid problems.
Abstract: As carbon dioxide emissions to the atmosphere continue from fossil fuel consumption, the Earth's climate will change considerably over the course of this century due to the resulting greenhouse effect. Societies and ecosystems will feel these global changes in the form of extreme weather, recurrent flooding and droughts, and other regional impacts. But how certain are we in projections about the distant future? Every prediction of a physical theory comes with error bars arising from limitations in data and approximations that are made for the sake of computational tractability. The scientific question is not whether the climate will change, but how likely it is to change by a given amount. This talk will discuss at a conceptual level the general mathematical formalism of statistical uncertainty quantification, and how it is being applied within the U.S. Department of Energy and other research institutions to estimate the plausible range of possible climate futures. The mathematical methods discussed may include Bayesian parameter estimation for nonlinear regression, Monte Carlo sampling, sensitivity analysis, model averaging, and surrogate modeling or emulation of expensive computer simulations. Applications may include probabilistic predictions of global warming and climate feedbacks, sea level rise from Antarctic ice sheet disintegration, and coastal flooding, among others. As time permits, I will also discuss educational training and career paths into this research field at the forefront of interdisciplinary science.
Abstract: This talk will introduce non-local mathematical models for two phase (fluid/fluid and fluid/solid) systems. We will discuss the mathematical basis of non-local models, and their role in modeling non-Newtonian fluids where local models cannot capture all effects. First, we propose a non-local model for surface tension obtained in the form of an integral of a molecular-force-like function added to the Navier-Stokes momentum conservation equation for multiphase fluid flows. We demonstrate that our model recovers both microscale and macroscale features of multiphase flow, eliminating the need for expensive hybrid MD-NS models, and provides strong advantages for modeling multiphase flows at length scales not feasible with MD simulations. Next, we extend the physics-informed neural network (PINN) method to learn a non-local viscosity model for a suspension of spherical particles using only velocity measurements. The PINN-inferred viscosity models agree with the empirical models for shear rates with large absolute values but deviate for shear rates near zero where the analytical models have an unphysical singularity.
Published article: http://dx.doi.org/doi:10.1016/j.jcp.2020.109732
The semester is about to end, and we are just as excited as you are! Join the SIAM Chapter for a Virtual Happy Hour on April 30th at 4:00 pm to celebrate the end of Finals. Food is on us if you RSVP early, so register now!
Please join the SIAM Student Chapter @ ASU in celebrating Black History Month with a Virtual Watch Party of Hidden Figures (2016) this Friday, Feb. 26th at 4:00 pm!
Want to learn how to successfully apply for jobs and polish your professional documents? Please join the SIAM Student Chapter @ ASU in a virtual Professional Development Workshop by Career and Professional Development Services (CPDS) on Tuesday, Nov. 17 at 10:00 am! Register here to participate. The zoom link will be provided upon registration. We look forward to seeing you there!