BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Michigan Institute for Computational Discovery and Engineering - ECPv6.17.1//NONSGML v1.0//EN
CALSCALE:GREGORIAN
METHOD:PUBLISH
X-WR-CALNAME:Michigan Institute for Computational Discovery and Engineering
X-ORIGINAL-URL:https://micde.umich.edu
X-WR-CALDESC:Events for Michigan Institute for Computational Discovery and Engineering
REFRESH-INTERVAL;VALUE=DURATION:PT1H
X-Robots-Tag:noindex
X-PUBLISHED-TTL:PT1H
BEGIN:VTIMEZONE
TZID:America/Detroit
BEGIN:DAYLIGHT
TZOFFSETFROM:-0500
TZOFFSETTO:-0400
TZNAME:EDT
DTSTART:20230312T070000
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:-0400
TZOFFSETTO:-0500
TZNAME:EST
DTSTART:20231105T060000
END:STANDARD
BEGIN:DAYLIGHT
TZOFFSETFROM:-0500
TZOFFSETTO:-0400
TZNAME:EDT
DTSTART:20240310T070000
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:-0400
TZOFFSETTO:-0500
TZNAME:EST
DTSTART:20241103T060000
END:STANDARD
BEGIN:DAYLIGHT
TZOFFSETFROM:-0500
TZOFFSETTO:-0400
TZNAME:EDT
DTSTART:20250309T070000
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:-0400
TZOFFSETTO:-0500
TZNAME:EST
DTSTART:20251102T060000
END:STANDARD
BEGIN:DAYLIGHT
TZOFFSETFROM:-0500
TZOFFSETTO:-0400
TZNAME:EDT
DTSTART:20260308T070000
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:-0400
TZOFFSETTO:-0500
TZNAME:EST
DTSTART:20261101T060000
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTART;TZID=America/Detroit:20250328T120000
DTEND;TZID=America/Detroit:20250328T130000
DTSTAMP:20250324T152136Z
CREATED:20250324T152136Z
LAST-MODIFIED:20250324T152136Z
UID:10000816-1743163200-1743166800@micde.umich.edu
SUMMARY:FSML Lecture Series - Lianghao Cao (Caltech): Derivative-Informed Operator Learning with Applications to Cost-Efficient Bayesian Inversion
DESCRIPTION:Zoom link \nBio: Dr. Lianghao Cao is a Postdoctoral Scholar Research Associate from the Department of Computing and Mathematical Sciences at the California Institute of Technology. He obtained a B.S. in Engineering Mechanics from the University of Illinois at Urbana-Champaign and a Ph.D. in Computational Science\, Engineering\, and Mathematics from The University of Texas at Austin. His research blends mechanistic modeling\, uncertainty quantification\, and scientific machine learning to understand\, enhance\, and control the quality\, validity\, and reliability of simulation-based predictions of complex physical systems. \nDerivative-Informed Operator Learning with Applications to Cost-Efficient Bayesian Inversion\nAbstract: This talk focuses on a derivative-informed supervised learning method for efficiently building machine learning surrogates of high-fidelity computational models\, particularly those governed by parametric partial differential equations. Unlike the conventional supervised learning method that treats the model as a black box\, our approach leverages additional model sensitivity information\, extracted via solving forward or adjoint sensitivity equations. This sensitivity information is integrated into the surrogate’s architecture and training process based on rigorous error analysis. We refer to such a surrogate construction as DINO (derivative-informed neural operator). \nDINO offers two key advantages over conventional surrogate construction. First\, it significantly improves the cost-accuracy trade-off for a wide range of models\, often by one to two orders of magnitude. Second\, it directly controls the surrogate Jacobian (Fréchet derivative) errors\, thus enhancing performance in surrogate-driven outer-loop problems that use gradient- and Hessian-based optimization algorithms. We demonstrate DINO’s capability to accelerate infinite-dimensional Bayesian inversion. First\, we show that geometric MCMC driven by DINO achieves a 2–9x speed up in asymptotically exact posterior sampling. Second\, we introduce LazyDINO\, a DINO-driven measure transport method for amortized Bayesian inversion\, which is one to two orders of magnitude more cost-efficient than competing methods.\nThis talk is based on joint work with Michael Brennan\, Joshua Chen\, Omar Ghattas\, Youssef Marzouk\, and Thomas O’Leary-Roseberry.
URL:https://micde.umich.edu/event/fsml-lecture-series-lianghao-cao-caltech-derivative-informed-operator-learning-with-applications-to-cost-efficient-bayesian-inversion/
LOCATION:2636 GGBA\, 2350 Hayward St\, Ann Arbor\, MI\, United States
CATEGORIES:Engineering,FSML,Science
ATTACH;FMTTYPE=image/png:https://micde.umich.edu/wp-content/uploads/2025/03/Lianghao-Cao-Caltech.png
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/Detroit:20250228T120000
DTEND;TZID=America/Detroit:20250228T130000
DTSTAMP:20250227T225237Z
CREATED:20250225T171723Z
LAST-MODIFIED:20250227T225237Z
UID:10000809-1740744000-1740747600@micde.umich.edu
SUMMARY:FSML Lecture Series - Doruk Aksoy: From Turbulent Flows to Video Games: Managing Large-Scale Data with Tensor Decomposition
DESCRIPTION:Zoom link \nBio:Doruk Aksoy is a 5th year PhD candidate in Aerospace Engineering and Scientific computing at the University of Michigan\, working under the supervision of Prof. Alex Gorodetsky. Prior to joining UM\, he studied Mechanical Engineering at Bogazici University in Istanbul Turkey. During his PhD\, he worked on developing incremental tensor decomposition algorithms to accelerate scientific machine learning through data reduction. \nFrom Turbulent Flows to Video Games: Managing Large-Scale Data with Tensor Decomposition\nAbstract: The rapid advancement of large-scale parallel computing created a surge of interest in developing high-fidelity digital twins for complex systems. However\, the computational demands for training these models are immense\, requiring vast amounts of data. As the spatial and temporal resolution of simulations increases\, even data storage becomes a critical bottleneck. This talk presents how low-rank tensor decomposition methods can be used to exploit the structure in large-scale data. We showcase a diverse array of applications\, from 3D turbulent Navier-Stokes simulations to Minecraft gameplay videos\, demonstrating the versatility and power of these techniques.
URL:https://micde.umich.edu/event/fsml-lecture-series-doruk-aksoy/
LOCATION:2636 GGBA\, 2350 Hayward St\, Ann Arbor\, MI\, United States
CATEGORIES:Engineering,FSML,Science
ATTACH;FMTTYPE=image/png:https://micde.umich.edu/wp-content/uploads/2023/09/Aksoy-Doruk.png
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/Detroit:20241206T120000
DTEND;TZID=America/Detroit:20241206T130000
DTSTAMP:20241204T144214Z
CREATED:20241011T181202Z
LAST-MODIFIED:20241204T144214Z
UID:10000779-1733486400-1733490000@micde.umich.edu
SUMMARY:FSML Lecture Series - Anoushka Bhutani: Foundation Model for Molecular Design
DESCRIPTION:Zoom link \nBio: Anoushka is a third-year PhD student in Prof. Venkat Viswanathan’s group at the University of Michigan. Her research interests include machine learning for materials design and electrochemical battery modeling. \nFoundation Model for Molecular Design\nAbstract: The paradigm of molecular machine learning for material screening has accelerated material development cycles\, improved efficiency\, and reduced costs. However\, current state-of-the-art molecular property prediction models still require labeled training data generated using wet-lab experiments or Density Functional Theory (DFT) calculations. Their utility is limited by the scarcity and heterogeneity of labeled materials datasets. Foundation models (FMs) offer a solution to this: these models use self-supervised pre-training strategies to leverage unlabeled datasets and learn representations of data that can be applied to downstream tasks. Large unlabeled datasets of billions of synthesizable molecules are readily available. Prior attempts to train FMs for molecular property prediction demonstrate promise; however\, equivariant geometric models trained using supervised learning are still more accurate. This can be attributed to the fact that foundation models are extremely expensive to train and can be difficult to interpret; they require huge computing budgets\, complex distributed computing techniques\, and extensive hyperparameter searches. Our work addresses these challenges on three fronts: (1) we have prototyped a scalable workflow for distributed training of molecular foundation models (2) we have trained large foundation models using this workflow which demonstrates state-of-the-art molecular property prediction capabilities across several benchmarks\, and (3) we have applied model interpretability strategies such as the attention visualization to shed insight on molecular structure relationships learn by the transformer. \n 
URL:https://micde.umich.edu/event/workshop-seminaranoushka-bhutani-foundation-model-for-molecular-design/
LOCATION:2636 GGBA\, 2350 Hayward St\, Ann Arbor\, MI\, United States
CATEGORIES:Computational Science,Engineering,FSML,Graduate School,Graduate Students,Michigan Engineering,Rackham,Research,Science
ATTACH;FMTTYPE=image/png:https://micde.umich.edu/wp-content/uploads/2024/10/Copy-of-MICDE-2022-2023-Fellowship-Portraits.png
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/Detroit:20241115T120000
DTEND;TZID=America/Detroit:20241115T130000
DTSTAMP:20260522T154425Z
CREATED:20241012T182154Z
LAST-MODIFIED:20260522T154425Z
UID:10000785-1731672000-1731675600@micde.umich.edu
SUMMARY:FSML Lecture Series - Hongfan Chen: Global Geomagnetic Perturbation Forecasting with Quantified Uncertainty using Deep Gaussian Process
DESCRIPTION:Zoom link \nBio: Hongfan Chen is a third-year PhD student in the Department of Mechanical Engineering at the University of Michigan. His research interests include data assimilation\, uncertainty quantification\, and machine learning applications in space weather. \nGlobal Geomagnetic Perturbation Forecasting with Quantified Uncertainty Using Deep Gaussian Process\nAbstract: Accurately predicting the horizontal component of the ground magnetic field perturbation (dBH)\, as a proxy for Geomagnetically Induced Currents (GICs)\, is crucial for estimating the impact of geomagnetic storms and remains a topic under active investigation. The current state-of-the-practice Geospace model is computationally expensive for fine-grid global simulations while existing machine learning methods consistently tend to underestimate dBH. Additionally\, these models either lack uncertainty quantification (UQ) or provide UQ that lacks calibration. In this work\, as part of the NextGen SWMF project funded by NSF\, we develop a data-driven\, grid-free global model using deep Gaussian process (DGP)\, a Bayesian non-parametric approach that forecasts the dBH for the full surface of Earth with calibrated uncertainty. The model uses solar wind measurements and the Dst index as input\, and it is trained based on ground magnetometer station data provided by SuperMAG over the period 1995-2022. The model’s predictions are evaluated based on the Heidke skill score (HSS) for a total of 22 geomagnetic storms in 2015. We further test the model on the 2024 May 10-12 storm. The results demonstrate that our model outperforms the state-of-the-art model\, with predictions exhibiting high accuracy in mid-latitudes and high-latitude regions in the northern hemisphere. \n 
URL:https://micde.umich.edu/event/lecture-discussionsciml-lecture-series-8/
LOCATION:2636 GGBA\, 2350 Hayward St\, Ann Arbor\, MI\, United States
CATEGORIES:Engineering,FSML,Science
ATTACH;FMTTYPE=image/jpeg:https://micde.umich.edu/wp-content/uploads/2024/10/Hongfan_Chen.jpg
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/Detroit:20241018T120000
DTEND;TZID=America/Detroit:20241018T130000
DTSTAMP:20241104T135643Z
CREATED:20241012T182153Z
LAST-MODIFIED:20241104T135643Z
UID:10000783-1729252800-1729256400@micde.umich.edu
SUMMARY:FSML Lecture Series: Domain decomposition and coupling data-driven models of fluid flows by Christopher Wentland\, Sandia National Labs
DESCRIPTION:Zoom link \n \nAbstract: Simulating complex physical systems often requires joining non-uniform subsystems\, which may be characterized by different geometries or mesh topologies. Coupling these separate subsystems often relies on time-intensive meshing workflows or empirical coupling models\, which may not generalize well across all operational regimes. The Schwarz alternating method proposes to overcome these issues\, establishing an effective domain decomposition framework that allows for the coupling of arbitrary geometries. This talk presents a brief history of Schwarz-based coupling work at Sandia National Laboratories\, along with recent work on combining the Schwarz alternating method with data-driven modeling approaches\, namely projection-based reduced order models (PROMs) and operator inference. This approach can generate surrogates that are capable of simulating advection-dominated fluid flows with higher accuracy and lower cost than comparable monolithic models\, aiding analysis in many-query applications such as uncertainty quantification and engineering design. Several nuances of the Schwarz algorithm and their impacts on model performance are explored\, specifically non-overlapping decompositions and PROM hyper-reduction under domain decomposition. A look into ongoing Schwarz coupling work at Sandia discusses existing challenges and efforts to apply this approach to relevant engineering problems.
URL:https://micde.umich.edu/event/lecture-discussionsciml-lecture-series-6/
LOCATION:2636 GGBA\, 2350 Hayward St\, Ann Arbor\, MI\, United States
CATEGORIES:Engineering,FSML,Science
END:VEVENT
END:VCALENDAR