Every year, the Michigan Institute for Computational Discovery & Engineering (MICDE) Catalyst Grants fund innovative research projects in computational science that combine elements of mathematics, computer science, and cyberinfrastructure. Topics of interest include, but are not limited to:

  • Computational science approaches, algorithms, frameworks, etc.
  • Emerging paradigms in computing (exascale computing, quantum computing, FPGA computing, etc.)
  • Applications in emerging areas (neuroscience, ecology, evolutionary biology, human-made complex systems, mobility etc.)
  • Extensions of traditional computational sciences to complex decision making (reinforcement learning, transfer learning, neuromorphic computing, etc.)
  • Artificial Intelligence informing and informed by science

Generic big data problems that do not fundamentally advance computational science algorithms are not suitable for MICDE Catalyst Grants. Priority will be given to high-impact projects with potential to eventually attract external funding. MICDE expects to fund 3-4 one-year projects per year at up to $100,000 each.

Funded projects are listed below. Calls for proposals will be listed on this page when they are opened.

MICDE Catalyst Grants Showcase

Watch presentations about some of the game-changing research supported by the Catalyst Grants program. Click on each image to open the YouTube link in a new tab. Speakers include: Robert Krasny (Mathematics),  Monica Valluri (Astronomy), Vikram Gavini (Mechanical Engineering), Stephen Smith (Ecology and Evolutionary Biology), Xun Huan (Mechanical Engineering), and Yulin Pan (Naval Architecture and Marine Engineering).

Integral equation based methods for scientific computing

Prof. Robert Krasny (Mathematics) presents several different approaches to the numerical solution of partial differential equations.

Probing the nature of dark matter by modeling the milky way

Prof. Valluri (Astronomy) presents a multi-pronged approach to understanding the nature of dark matter with new dynamical tools and simulations.

Long time-scale simulations using exponential time-propagators

Prof. Gavini (ME) presents the ideas his team has been exploring to enable long time-scale simulations on a class of time-dependent problems.

Emergence of Bio. Complexity & Evolutionary Innovation in Plant Genomes

Prof. Smith (EEB) talks about the progress his team has made in developing new genomic & trait detection methods.

Towards Bayesian UQ in DL Models for Brain Tumor Segmentation

Prof. Huan (ME) explains how his team is tackling uncertainty in deep learning predictions in medical applications.

Real-time phase resolved ocean wave forecast

Real-time phase-resolved prediction of ocean waves is crucial for the safety of offshore operations. Prof. Yulin Pan (NAME) and his team explain how they are tackling this problem.

2022 Funded Projects

MICDE Catalyst Grants enables junior faculty innovations in computational science

Four projects lead by outstanding junior faculty have been awarded the 2022 MICDE Catalyst Grant funding. “Ranging from computer vision-enabled insights to animal behavior, through fundamental mathematical frameworks to advance quantum physics and materials discovery, to probability-based computational modeling of material behavior, the 2022-23 cohort of catalyst grant awardees will continue MICDE’s tradition of uncovering new frontiers of computational science”, said MICDE Director Krishna Garikipati, Professor of Mechanical Engineering, College of Engineering, and Professor of Mathematics, LSA.

Evidential Crystal-Graph Convolutional Neural Networks for Efficient Global Optimization of Electrocatalysts

Researchers: Bryan Goldsmith (U-M, Chemical Engineering), Suljo Linic (U-M, Chemical Engineering)

Description: Trial-and-error experimental approaches to catalyst discovery are time consuming and expensive. Computational screening of catalysts using quantum mechanical methods such as density functional theory (DFT) modeling are highly useful, but these approaches are still limited because of the high computational cost. Machine learning (ML) using deep neural networks has emerged as a powerful complement to high-throughput screening, enabling prediction of catalysts at much faster speeds compared to using only experimental methods or DFT modeling. However, the full utility of ML models to explore large catalyst spaces can only be obtained if it is straightforward to identify when model predictions are very uncertain or accurate, Fig. 1. Established uncertainty quantification approaches for neural networks are typically costly to obtain and have limitations in evaluating prediction errors for catalyst space exploration. To enable broader use of ML for catalyst discovery, new uncertainty quantification approaches must be developed. We will implement evidential regression with crystal-graph convolutional neural networks (CGCNN) to enable accurate prediction of model uncertainty and accelerate electrocatalyst optimization for energy applications. We will use evidential regression CGCNN within an optimization framework to discover electrocatalysts for sustainable fuel generation, which is critical to combat climate change.

Fig. 1. Illustration of model uncertainty estimation. It is essential to know when the neural network predictions are low uncertainty or high uncertainty as this knowledge is needed for uncertainty-guided learning for sample-efficient model training and for enhanced catalyst optimization. Image adapted from Oleimany, A. P.; Amini, A.; Goldman, S.; Rus, D.; Bhatia, S. N.; Coley, C. W. ACS Cent. Sci. 2021, 7 (8), 1356–1367.

Computer Vision Tools for Automatic Animal Behavioral Classification in Complex
Environments

Researchers: Ada Eban-Rothschild (U-M, Psychology), Justin Johnson (Electrical Engineering, Computer Science)

Description: Our research team will develop an accessible computer vision toolbox to automatically track multiple animals and classify their behaviors in complex social environments. We will harness state-of-the-art developments in machine learning and computer vision as well as a rich dataset of manually annotated video recordings.  

 

Our proposed Behavior classification pipeline. We will use three steps: (1) Preprocessing our video data by creating a localized video stream using our SOT or MOT pipelines. The example illustrates three tracked animals. (2) Encoding each animal video stream into a latent representation. (3) Classify each animals’ latent representation into our behavioral categories using a classifier. Our 2nd and 3rd steps are jointly optimized using our behavior classification dataset.

Multi-scale Continuous Tensor Networks for Quantum Simulations

Researcher: Alex Gorodetsky (U-M, Aerospace Engineering)

Description: Methods for efficient simulations of quantum many-body problems are essential for theoretical studies of physical and chemical systems where quantum effects are important. These simulations either take the form of solving the high-dimensional Schrodinger Equation for wave functions that represent all information about the system or as high-dimensional optimization problems whose solutions represent the energy of the system. This work will develop faster scalable computational tools to solve higher dimensional quantum many-body problems. The novelty of these tools lies in their ability to simultaneously exploit low-rank and locally multi-scale structure via continuous tensor networks.

Hierarchical Tucker decomposition extracts multi- scale structure

Probability Mechanisms Map of Dislocation-Obstacle Interaction as an Enabler of
Physics-based Multiscale Modeling on Precipitation Hardening

Researchers: Yue Fan, (U-M, Mechanical Engineering), Xun Huan (U-M, Mechanical Engineeering)

Description: Structural materials’ mechanical properties are largely controlled by the evolutions and interactions of their inside microstructural features called defects. In particular, the interaction between line defects (known as dislocations) and other obstacles (e.g. impurity precipitates) is playing a decisive role. Multiscale materials modeling (MMM), as a widely adopted strategy over the past few decades, has provided unprecedented details on defect’s evolutions and interactions from the atomic level. However, in the current MMM paradigm, significant gap still exists on transferring fundamental mechanisms of local dislocation-obstacle interaction into a predictive global constitutive relationship. The objective of this proposal is to develop a novel modeling framework to probe the accessible transition pathways and uncover the competing atomistic interaction mechanisms for any given dislocation-obstacle pair. The project will establish a capability of quantifying the occurrence probability of each mechanism and its embedded uncertainty over broad thermo-mechanical parameter space covering the realistic timescales.

A probabilistic atomic-to-mesoscale modeling interface with be developed through a Monte Carlo process, by utilizing a probability mechanisms map

“This year’s MICDE Catalyst Grants bolster junior faculty innovations in computational science”

Krishna Garikipati
Director, MICDE

2021 Funded Projects

MICDE Catalyst Grants continue to forge new fronts in computational science

Five broad and high impact computational science projects have been awarded a 2021 MICDE Catalyst Grant funding. “The 2021-22 cohort of MICDE  Catalyst Grants range from quantum computing for engineering science, AI for the physics of cancer, and computational advances in hazards engineering, through mathematical advances in data science, and bioengineering. These projects represent new frontiers of computational research spearheaded by the Institute through its initiatives”, said MICDE Director Krishna Garikipati, Professor of Mechanical Engineering, College of Engineering, and Professor of Mathematics, LSA.

Scalable Inference of Spatially-varying Graphical Models with Applications in Genomics

Researchers: Salar Fattahi (U-M, Industrial and Operations Engineering), Arvind Rao (U-M, Computational Medicine & Bioinformatics, Radiation Oncology)

Description: Contemporary systems are comprised of massive numbers of interconnected components that interact according to a hierarchy of complex, dynamic, and unknown topologies. The unknown and varying nature of these systems necessitates the development of efficient inference methods for these STGM. A popular approach to achieve this goal is based on the so-called maximum-likelihood estimation (MLE), however, these theoretically powerful MLE-based methods suffer from fundamental drawbacks rendering them impractical in realistic settings.

With the goal of bridging this knowledge gap, this project aims to revisit the standard MLE as the “Holy Grail” of the inference methods for graphical models, and precisely pinpoint and remedy the scenarios where it breaks down. If successful, this project will be the first systematic inference framework that can achieve the best of both worlds, computational efficiency and favorable statistical performance, in a unified fashion.

Spatially-informed clusters from spatial transcriptomics data of human GBM tis-
sue sample.

Cancer Cells: Greedy Individuals or Team Players?

Researchers: Gary Luker (U-M, Radiology, Microbiology and Immunology, Biomedical Engineering), Nikola Banovic (U-M, Electrical Engineering and Computer Science), Xun Huan (U-M, Mechanical Engineeering), Jennifer Linderman (U-M, Biomedical Engineering, Chemical Engineering), Kathryn Luker (U-M, Radiology)

Description: The immediate goal of this project is to develop a physics/chemistry-aware inverse reinforcement learning (IRL) computational framework to discover how heterogeneous cancer cells function singly or collectively to drive cancer progression. The long-term goal of this research centers on understanding single-cell and cooperative decision-making that drive tumor growth, metastasis, and recurrence. The proposed work is computational science in nature, developing new and scalable artificial intelligence (AI) algorithms that leverage cell imaging data to extract knowledge on cancer cell behavior and predict interventions.

Examples of quantitative imaging data for computational modeling of cancer cell heterogeneity and teamwork.

“These projects represent new frontiers of computational research spearheaded by the Institute through its initiatives”

Krishna Garikipati
Director, MICDE

Fast Linear Algebra in the Noisy Intermediate-scale Quantum Era

Researcher: Shravan Veerapaneni (U-M, Mathematics)

Description: Recent impressive progress in quantum technology, particularly in programmable quantum computers, has invigorated a renewed interest in quantum algorithm research. This project aims to develop quantum and quantum-inspired solvers for linear systems appearing in scientific computing (such as discretized partial differential equations). The primary mode of solution is using Rayleigh quotient reformulation and applying variational quantum Monte Carlo (VQMC). In addition to providing a toolkit for performing high-dimensional linear algebra, which is of intrinsic interest, the proposed solver provides a quantum-inspired classical benchmark for assessing the quantum computational advantage of the recently developed variational quantum linear solver.

Computational Modeling of Household Level Damage and Resilience Based on Location Data

Researchers: Seth Guikema (U-M, Industrial & Operations Engineering, Civil & Environmental Engineering), Jeremy Bricker (U-M, Civil & Environmental Engineering)

Description: A critical aspect of community resilience to disasters that has not been addressed by traditional coarse-scale resilience engineering is within-community inequities in resilience – who is resilient to what? 

The goal of this project is to develop an integrated approach for assessing household-level resilience and inequities in resilience during coastal flooding events, specifically by improving building-level flood and fragility estimates for coastal flooding events, and developing a new approach for estimating what essential services are the main constraints on individuals returning to a more normal life post-hazard and assess inequities in resilience to coastal flooding events. 

Maximum water depth as a function of model resolution for a hypothetical flood in the Netherlands, from Brusseeet al. (2021).

Next Generation Computational Tools for Particle-laden Biological Flows in Subject-specific Geometries

Researchers: Jesse Capecelatro (U-M, Mechanical Engineering, Aerospace Engineering), Alberto Figueroa (U-M, Biomedical Engineering, Vascular Surgery)

Description: Fluid mechanics plays a crucial role in many physiological processes on health and disease. Given recent advances in medical imaging, computational power, and mathematical algorithms, real-time patient-specific computational fluid dynamics is now becoming possible. Yet, many problems involve complex interactions between fluid and biological particles in which existing models are either too expensive to simulate at full scale or unable to properly capture important hydrodynamics taking place at the smallest scales. This project will develop a versatile and massively parallel framework to bridge this gap. The numerical framework will be designed to simulate a large number of particles within the human body. This will help better understand cardiovascular diseases, from stroke, to rigid calcite particles in the ear canal responsible for vertigo.

Example of an image-based geometric model of a human aorta, discretized using an unstructured linear tetrahedral mesh.

2020 Funded Projects

Four U-M Projects Aim to Advance Computational Discovery

Research teams from across the University of Michigan will share $240,000 in awards to explore projects ranging from drug discovery and galactic formation to bacterial colonies and turbulence. “These four projects have the potential to catalyze and reorient the directions of their research fields by developing and harnessing powerful paradigms of computational science,” said MICDE Director Krishna Garikipati, professor of mechanical engineering, College of Engineering, and professor of mathematics, LSA.

Read more in the University Record.

Accelerated Computation of Resolvent Modes for High-Dimensional Dynamical Systems

Researchers: Aaron Towne (Mechanical Engineering)

Description: Turbulence, the disorganized motion of a fluid (often associated with a bumpy ride on an airplane), is ubiquitous in science and engineering, impacting everything from the flight of a golf ball, to fuel efficiency and noise of an engine, to the formation of stars.  One important tool for studying and modeling turbulent flows is a mathematical framework called resolvent analysis, which identifies energy amplification mechanisms key to generating and sustaining turbulence.  Unfortunately, resolvent analysis requires significant computational resources when applied to realistic engineering systems.  The goal of this project is to develop a new algorithm that reduce the cost of resolvent analysis of large systems by several orders of magnitude.  This capability could lead to a better theoretical understanding of turbulence and improved design of engineering systems involving turbulent flow. Read more.

Top: Pressure (grayscale) and vorticity (colors) in a turbulent jet. Bottom: Resolvent analysis provides an excellent model of the dominant structures within the pressure field. Reducing the cost of computing these models could lead to quieter airplanes in the future, among other engineering objectives.

Uncovering the Origins of the Local Group of Galaxies with Tailored Initial Conditions

Researcher: Oleg Gnedin (Astronomy)

Description: This project aims to uncover the origin of the local group of galaxies by developing a novel technique for custom tailoring initial conditions to simulate a specific chosen region of the universe. The computational challenges of this technique are to enforce the continuity of physical variables and consistency of simulated outcomes, and to improve computational performance of galaxy formation codes. This technique could benefit all galaxy formation codes used by different groups around the world. Read more.

Exploring Spatiotemporal Biofilm Development Through the Computational Looking Glass 

Researchers: Alexander Rickard (Epidemiology), Marisa Eisenberg (Epidemiology, Complex Systems, Mathematics)

Description: Biofilms are architecturally complex assemblages of microbial cells that form on surfaces that are on and within our bodies and exist on almost every natural and man-made surface. Differences in biofilm architecture will alter how constituent cells dynamically interact with each other and how they interact with their environment. This project will focus on the computational analysis of time-lapse collected image stacks (i.e. optical sections) of single-species biofilms to study their architectural development. Read more.

 

Preliminary data showing that biofilms form complex architectures over time.

Leveraging Generative Artificial Intelligence for the De Novo Design of Biomolecular Probes

Researcher: Aaron Frank (Chemistry)

Description: Structure-based drug discovery involves exploring chemical space in search of novel compounds that are likely to bind to and modulate the activity of a biomolecular drug target. There is an urgent need for efficient strategies for exploring chemical space, conditioned by the target’s unique biophysical properties. In this project, researchers will use a structure-aware approach that combines generative artificial intelligence models and molecular docking to rapidly explore chemical space and generate target-specific virtual libraries. Such target-specific virtual libraries will likely contain compounds that medicinal chemists can use as starting points for developing novel drug candidates.

The project’s “sample-and-dock” approach to generate target-specific screening library for drug targets.

2019 Funded Projects

Determining the 3D shape of Milky Way’s Dark Matter Halo

Researchers: Monica Valluri (Astronomy), August Evrard (Physics), Kohei Hattori (Astronomy), Eugene Vasiliev (University of Cambridge), Pablo Fernández de Salas (University of Stockholm), and Katherine Freese (UT Texas and University of Stockholm)

Description: The objectives of this project are to use data from the European Space Agency’s Gaia satellite to determine the shape of the Milky Way’s dark matter halo and how it this shape changes with distance from the center. The background image shows Gaia’s multi-color map of the Milky Way and two nearby satellite galaxies, the Large and Small Magellanic Clouds (bottom right).  This image is not a photograph but a map comprised of billions of individual stars in the Milky Way and its neighborhood. For this project the team will use data obtained by Gaia that gives positions and velocities for stars in the Milky Way’s halo along with the new computational modeling tools they are developing. They will use the fact that most halo stars travel through the halo on regular trajectories (like the one pictured by the blue curve) to compute conserved quantities (“actions”) for each of hundreds of thousands of stars. Modeling the motions of halo stars will allow the team to set constrains on the density profile and the shape of the Milky Way’s dark matter halo. By comparing their derived dark matter distribution with results from simulations they hope to shed light on the nature of the elusive dark matter particle.

The background image is a multi-color image of the Milky Way disk, its halo and nearby satellite galaxies obtained with the European Space Agency’s Gaia Satellite (http://sci.esa.int/gaia/) . The blue curve shows an example of (half) of a regular trajectory that a star in the halo of the Milky Way might follow

Enabling Tractable Uncertainty Quantification for  High-Dimensional Predictive AI Systems in Computational Medicine

Researchers: Xun Huan (Mechanical Engineering), and Arvind Rao (Computational Medicine and Bioinformatics and Radiation Oncology )

Description:  Artificial intelligence (AI) systems are powerful tools in healthcare and medicine. However, it is crucial to understand how much one can trust the AI analyses and predictions, especially when adopting them for decision-making where inappropriate choices may result in dire consequences.  In this project, we start by developing the computational and algorithmic foundations for performing uncertainty quantification (UQ) in machine learning (ML) models. We tackle this by creating new computational methods and leveraging high-performance computing, to capture and construct uncertainty distributions for high dimensional deep neural networks (of tens of millions of weight parameters). We focus on medical AI models used for detecting IDH (isocitrate dehydrogenase) gene mutation from MRI (magnetic resonance imagine) brain tumor images. The resulting product will be ML models that produce not only a single output, but a spread of predictions which also reflects its predictive quality and uncertainty.

The DNN model takes in MRI image and predicts the probability of IDH gene mutation. e.g. while predicted probability is 80%, the uncertainty surrounding this estimate is typically never reported.

Real-Time Phase-Resolved Ocean Wave Forecast with Data Assimilation Enabled by GPU-accelerated Computation

Researcher: Yulin Pan (Naval Architecture and Marine Engineering)

Description: The purpose of this project is to develop a new computational framework for the onboard real-time forecast of phase-resolved ocean wave field with data assimilation capability. The prediction and assimilation algorithms will be accelerated on a CPU-GPU hardware architecture for real-time applications. Integrated with the remote wave sensing technology, this work will lead to enhanced safety, efficiency and autonomy in marine operations.

Envisaged application in autonomous path planning.

Long time-scale simulations using exponential time-propagators 

Researchers: Vikram Gavini (Mechanical Engineering)

Description:  This effort is aimed at developing scalable and efficient algorithms for long-time scale simulations of dynamical phenomena in materials. The approaches to be developed are expected to enhance our ability to study a wide range of time-dependent phenomena from electron dynamics to elastic response of materials.

Algorithmic solutions to manage power consumption on exascale systems

Researchers: Eric Johnsen (Mechanical Engineering), Henry Hoffman (University of Chicago), and Jeffrey Hittinger (Lawrence Livermore National Lab)

Description: Our objective is to develop a quantitative strategy for power management at the exascale, given a desired solution accuracy. For this purpose, our approach integrates high-order methods development with mixed-precision computing, lossy data compression, and applications monitoring power consumption.

Convergence rate of Recovery DG vs. BR2 (the current gold standard) for linear diffusion on a Cartesian mesh. For the same p=2, Recovery achieves 8th order for Recovery, while BR2 is 3rd order.

Simulation-based discovery of robust algorithms for targeting of infectious disease screening and intervention

Researcher: Jon Zelner (Epidemiology), and Seth Guikema (Industrial and Operations Engineering and Civil and Environmental Engineering)

Description: In settings with a high burden of infectious diseases, such as Tuberculosis (TB), there is a growing need for tools that can help public health professionals find and treat cases more quickly and effectively. In this project, we will be continuing development of a spatiotemporal simulation model of coupled household and community TB in a high-incidence setting. In our initial work on this project, we have used this model to compare different intervention scenarios that take advantage of information on the spatial location and contact networks of TB cases to target interventions. In the next phase of this project, we plan to use this simulation platform as a tool for developing adaptive interventions that can respond to changing epidemiological conditions, i.e. a sharp rise in incidence indicative of an outbreak, and modify screening and intervention strategies to deal with the types of heterogeneity that makes tackling real-world infectious disease problems highly challenging.

Spatial hotspot of strongly elevated multi-drug resistant (MDR-TB) incidence in Lima, Peru illustrated in red. (Figure from Zelner et. al., JID 2016 [2])

Related Publications

J. Havumaki ,T. Cohen, C. Zhai, J. C. Miller,S. D. Guikema, M. C. Eisenberg, J. Zelner. “Protective impacts of household-based tuberculosis contact tracing are robust across endemic incidence levels and community contact patterns.” PLoS computational biology vol. 17, 2 e1008713. 8 Feb. 2021, doi:10.1371/journal.pcbi.1008713

Nina B. Masters, Marisa C. Eisenberg, Paul L. Delamater, Matthew Kay, Matthew L. Boulton, Jon Zelner. “Fine-scale spatial clustering of measles nonvaccination that increases outbreak potential is obscured by aggregated reporting data.” Proceedings of the National Academy of Sciences Nov 2020, 117 (45) 28506-28514; DOI: 10.1073/pnas.2011529117

Hierarchical computing for dynamic evolutionary inference of complexity

Researchers: Stephen Smith (Ecology and Evolutionary Biology)

Description:  We propose to develop new tools grounded in emerging techniques for accommodating heterogeneity in genomic and trait datasets to enable statistical comparison of the formation and evolution of modules across many taxa that efficiently handle multivariate datasets consisting of multiple sources including morphology, genomes, biochemical data, and gene expression. These new methods will identify common patterns of evolutionary rate and mode across multiple genes and traits, allow for lineage specific heterogeneity, scale to the large dimensions common in modern datasets, and break the false dichotomy of genes vs traits. Given the size of the datasets, and the necessity to explore model complexity, these methods will require significant computation and a hierarchical computational approach: distributed computing for independent analyses (e.g., individual genes), multi-core parallel computing of individual analyses where likelihoods are calculated in parallel on shared memory resources, and GPU computing for more extensive model explorations that require extensive matrix calculations. The methodological developments will be implemented in, gophy, a package developed by the participants of the proposal.

2018 Funded Projects

Exploring Quantum Embedding Methods for Quantum Computing

Researchers: Emanuel Gull, Physics; Dominika Zgid, Chemistry

Description:  The research team will design quantum embedding algorithms that can be early adopters of quantum computers on development of advanced materials for possible applications in modern batteries, next-generation oxide electronics, or high-temperature superconducting power cables.

Teaching Autonomous Soft Machines to Swim

Researchers:  Silas Alben, Mathematics; Robert Deegan, Physics, Alex Gorodetsky, Aerospace Engineering
Description:  Self-oscillating gels are polymeric materials that change shape, driven by chemical reactions occurring entirely within the gel. The research team will develop a computational and machine learning program to discover how to configure self-oscillating gels so that they undergo deformations that result in swimming. The long term goal is to develop a general framework for controlling autonomous soft machines.

Caption: Chemical waves and elastic deformations in thin gel sheets.
(a & b) Target and spiral waves in the BZ (Belousov-Zhabotinsky) chemical reaction.
(c) Numerical simulations of the BZ reaction. (d) Resulting reference metrics and shapes for the gel sheet.(e) Experimental images of oscillating gels.

Urban Flood Modeling at “Human Action” Scale: Harnessing the Power of Reduced-Order Approaches and Uncertainty Quantification

Researchers:  Valeriy Ivanov, Civil and Environmental Engineering; Nick Katopodes, Civil and Environmental Engineering; Khachik Sargsyan, Sandia National Labs

Description:  The research team will enhance urban flood monitoring and prediction using NASA Cyclone Global Navigation Satellite System (CYGNSS) data, taking advantage of state-of-the-science uncertainty quantification tools in a proof-of-concept urban flooding problem of high complexity.

The problem of flood prediction incurs multiple uncertainties (a) and is of high computational complexity (b). New remote sensing data from the CYGNSS mission (c) can inform complex physically-based simulations (d) but in order to achieve feasibility of real-time solutions and uncertainty quantification, novel approaches are required. This project will develop reduced-order modeling tools (e) as innovative, parsimonious representation of rigorous hydrologic and hydrodynamic model formulations to efficiently obtain probability density distributions of one or many quantities of interest (f).

Deciphering the meaning of human brain rhythms using novel algorithms and massive, rare datasets

Researcher: Omar Ahmed, Psychology, Neuroscience and Biomedical Engineering

Description: The team will develop a set of algorithms for use on high performance computers to analyze de-identified brain data from patients in order to better understand what electrical oscillations tell us about rapidly changing behavioral and pathological brain states.

Polar plots showing the rhythmic phases of spikes fired by human neurons, revealing systematic variations across space and time.

Advancing the Computational Frontiers of Solution-Adaptive, Scale-Aware Climate Models

Researchers:  Christiane Jablonowski, Climate and Space Sciences and Engineering; Hans Johansen, Lawrence Berkeley National Lab

Description:  Researchers will further develop a 3-D mesh adaptation model for climate modeling, allowing computational resources to be focused on phenomena of interest such as tropical cyclones or other extreme weather events. The project will also introduce data-driven machine learning paradigms into modeling of clouds and precipitation.

Structure of a block-structured adaptive grid that overlays a so-called ‘cubed-sphere’ base grid. The atmospheric vortices (in color) resemble idealized tropical cyclones that are captured by the refined grid patches. The adaptations are guided by the location and strength of the rotational motion.

Embedded Machine Learning Systems To Sense and Understand Pollinator Behavior

Researchers: Robert Dick, Electrical Engineering and Computer Science; Fernanda Valdovinos Ecology and Evolutionary Biology, Center for Complex Systems; Paul Glaum, Ecology and Evolutionary Biology

Description: To understand the mechanisms driving the population dynamics of pollinators, the research team will develop technologies for deeply embedded hardware/software learning systems capable of remote, long term, autonomous operation; and will analyze the resulting new data to better understand pollinator activity.

Embedded sensing and machine learning to distinguish pollinators (Bombus impatiens pictured) from other sound sources in natural environments.

Deep Learning for Phylogenetic Inference

Researcher: Jianzhi Zhang, Ecology and Evolutionary Biology; Yuanfang Guan, Computational Medicine and Bioinformatics

Description: The research team will use deep neural networks to infer molecular phylogenies and extract phylogenetically useful patterns from amino acid or nucleotide sequences, which will help understand evolutionary mechanisms and build evolutionary models for a variety of analyses.

Molecular sequence data are input into layers of neural network structures to produce a score for each possible tree topology describing how well the tree fits the data.

2017 Funded Projects

From Spiking Patterns to Memory formation — Tools for Analysis and Modeling of Network-wide Cognitive Dynamics of the Brain

Researchers: Sara Aton, Department of Molecular, Cellular and Developmental Biology and Michal Zochowski, Department of Physics, Biophysics Program

Description: The aim of the research is to develop models as well as analysis tools to understand network-wide spatio-temporal patterning underlying experimentally observed neural spiking activity. The research team has developed novel tools to analyze dynamics of neuronal representations across time, before during and after learning. These tools, for the first time, compare the stability of network dynamics before and after memory encoding.

Integral Equation Based Methods for Scientific Computing

Researcher: Robert Krasny, Department of Mathematics

Description: This project expands the application of numerical methods in which the differential equation is first converted into an integral equation by convolution with the Green’s function, followed by discretization and linear solution. Recent advances in numerical analysis and computing resources make this expansion possible, and the research team believes that integral equation-based numerical methods are superior to traditional methods in both serial and parallel computations. The project will attempt to apply these numerical methods to studies of viscous fluid flow, protein/solvent electrostatics, and electronic structure.

Computational Energy Systems

Researchers: Pascal Van Hentenryck, Industrial and Operations Engineering (IOE); E. Byon, IOE; R. Jiang, IOE; J. Lee, IOE; and J. Mathieu, Electrical Engineering and Computer Science

Description: The research team aims to develop new algorithms for the U.S. electrical power grid that integrate renewable energy sources, electrification of transportation systems, the increasing frequency of extreme weather events, and other emerging contingencies.

Black Swans, Dragon Kings, and the Science of Rare Events: Problems for the Exascale Era and Beyond

Researchers: Venkat Raman, Aerospace Engineering; Jacqueline Chen, Sandia National Laboratory; and Ramanan Sankaran, Oak Ridge National Laboratory.

Description: The purpose of the project is to develop the computational frameworks for exploring the tails of distributions, which lead to rare but consequential (and often catastrophic) outcomes. Two such rare events are “Black Swans” (occurring from pre-existing but unencountered events) and “Dragon Kings (occurring due to an external shock to the system). The methods developed are expected to have application in aerospace sciences, power generation and utilization, chemical processing, weather prediction, computational chemistry, and other fields.