Applied/ACMS: Difference between revisions

From UW-Math Wiki
Jump to navigation Jump to search
← Older edit
VisualWikitext
Spagnolie (talk | contribs)
1,345 edits
Spagnolie (talk | contribs)
1,345 edits
 
(3 intermediate revisions by the same user not shown)
Line 19: Line 19:
|Sep 19*
|Sep 19*
|[https://www.anl.gov/profile/zichao-di Zichao (Wendy) Di] (Argonne National Laboratory)
|[https://www.anl.gov/profile/zichao-di Zichao (Wendy) Di] (Argonne National Laboratory)
|Multimodal Inverse Problems and Multilevel Optimization for X-ray Imaging Science
|[[#Di|Multimodal Inverse Problems and Multilevel Optimization for X-ray Imaging Science]]
|Rycroft/Li
|Rycroft/Li
|-
|-
|Sep 26
|Sep 26
|[https://scholar.google.com/citations?user=Imuw5CMAAAAJ&hl=en&oi=ao Pouria Behnoudfar] (UW)
|[https://scholar.google.com/citations?user=Imuw5CMAAAAJ&hl=en&oi=ao Pouria Behnoudfar] (UW)
|Bridging Conceptual and Operational Models: A Physics-Guided Machine Learning Framework for Enhanced Climate Simulation
|[[#Behnoudfar|Bridging Conceptual and Operational Models: A Physics-Guided Machine Learning Framework for Enhanced Climate Simulation]]
|Spagnolie
|Spagnolie
|-
|-
Line 73: Line 73:
|-
|-
|Dec 5
|Dec 5
|[https://mesomod.weebly.com/ Jiamian Hu] (UW; Engineering)
|[https://mesomod.weebly.com/ Jiamian Hu] (UW)
|TBD
|TBD
|Chen
|Chen
Line 86: Line 86:
==Abstract==
==Abstract==


<div id="Chandler">
<div id="Di">
'''Zichao (Wendy) Di (Argonne National Laboratory)'''
'''Zichao (Wendy) Di (Argonne National Laboratory)'''


Line 95: Line 95:
A second line of work focuses on the computational challenge: even for single-modality inverse problems, the resulting optimization problems are large-scale, nonlinear, and nonconvex. Here, I will discuss multilevel optimization and stochastic sampling strategies that accelerate convergence by exploiting hierarchical structure in both parameter and data spaces.
A second line of work focuses on the computational challenge: even for single-modality inverse problems, the resulting optimization problems are large-scale, nonlinear, and nonconvex. Here, I will discuss multilevel optimization and stochastic sampling strategies that accelerate convergence by exploiting hierarchical structure in both parameter and data spaces.


Although developed separately, these two directions point toward a common goal: building scalable, optimization-based frameworks that make the best use of diverse data to enable new discoveries in X-ray imaging science.<div id="Fraser"><div id="Luedtke"><div id="Zhdankin"><div id="Boffi"><div id="Shankar"><div id="Loevbak">
Although developed separately, these two directions point toward a common goal: building scalable, optimization-based frameworks that make the best use of diverse data to enable new discoveries in X-ray imaging science.
<div id="Lu"><div id="Vogman"><div id="Cockburn">
 
<div id="Behnoudfar">
'''Pouria Behnoudfar (UW Madison)'''
 
Title: Bridging Conceptual and Operational Models: A Physics-Guided Machine Learning Framework for Enhanced Climate Simulation
 
Operational models are high-resolution and contain many crucial variables. Despite numerous successes, biases persist in most of these models, particularly in identifying extreme events and accurately reproducing statistics. However, it is challenging to directly modify these models to improve their performance. Conceptual and intermediate coupled models accurately characterize certain features of nature. Yet, these coarse-grained models contain only a subset of variables within a specific domain. By leveraging the strengths of models of varying complexities, we develop a robust physics-driven machine learning modeling framework as the next-generation climate emulator, which bridges the model hierarchy through effective latent space data assimilation. The resulting bridging model not only inherits the benefits of the operational models, including their high resolution and comprehensive set of variables, but also globally enhances the accuracy through local dynamical and statistical improvements provided by the simpler models.  Our developed latent space technique identifies the dominant nonlinear features, which facilitates effective communication between models. The machine learning representation of the bridging model significantly enhances simulation efficiency, providing high-quality datasets and advancing the quantification of uncertainty in extreme events. The framework is applied to improve the performance of CMIP6 simulations in characterizing ENSO complexity by exploiting simpler yet statistically accurate models. 
 
== Archived semesters ==
== Archived semesters ==


Latest revision as of 03:51, 9 September 2025


Applied and Computational Mathematics Seminar


Fall 2025

Date Speaker Title Host(s)
Sep 19* Zichao (Wendy) Di (Argonne National Laboratory) Multimodal Inverse Problems and Multilevel Optimization for X-ray Imaging Science Rycroft/Li
Sep 26 Pouria Behnoudfar (UW) Bridging Conceptual and Operational Models: A Physics-Guided Machine Learning Framework for Enhanced Climate Simulation Spagnolie
Oct 3
Oct 10* Alexandria Volkening (Purdue) TBD Rycroft
Oct 17* Nick Derr (UW) TBD Spagnolie
Oct 24 Mike O'Neil (Courant) TBD Spagnolie
Oct 31 Hyukpyo Hong (UW) TBD Spagnolie
Nov 7* John Bush (MIT) TBD Spagnolie
Nov 14 Yukun Yue (UW) TBD Spagnolie
Nov 21* Jessie Levillain (CNES/INSA Toulouse) TBD Ohm
Nov 28 Thanksgiving
Dec 5 Jiamian Hu (UW) TBD Chen
Dec 12 Thomas Fai (Brandeis) TBD Rycroft

[Dates marked with an asterisk are close to weekends with a home game for the UW Badgers football team. Hotel availability around these dates is often limited if booked on short notice.]

Abstract

Zichao (Wendy) Di (Argonne National Laboratory)

Title: Multimodal Inverse Problems and Multilevel Optimization for X-ray Imaging Science

X-ray imaging experiments generate vast datasets that are often incomplete or ill-posed when considered in isolation. One way forward is multimodal data analysis, where complementary measurement modalities are fused to reduce ambiguity and improve reconstructions. A key question, both mathematically and practically, is how to identify which modalities to combine and how best to integrate them within an inverse problem framework.

A second line of work focuses on the computational challenge: even for single-modality inverse problems, the resulting optimization problems are large-scale, nonlinear, and nonconvex. Here, I will discuss multilevel optimization and stochastic sampling strategies that accelerate convergence by exploiting hierarchical structure in both parameter and data spaces.

Although developed separately, these two directions point toward a common goal: building scalable, optimization-based frameworks that make the best use of diverse data to enable new discoveries in X-ray imaging science.

Pouria Behnoudfar (UW Madison)

Title: Bridging Conceptual and Operational Models: A Physics-Guided Machine Learning Framework for Enhanced Climate Simulation

Operational models are high-resolution and contain many crucial variables. Despite numerous successes, biases persist in most of these models, particularly in identifying extreme events and accurately reproducing statistics. However, it is challenging to directly modify these models to improve their performance. Conceptual and intermediate coupled models accurately characterize certain features of nature. Yet, these coarse-grained models contain only a subset of variables within a specific domain. By leveraging the strengths of models of varying complexities, we develop a robust physics-driven machine learning modeling framework as the next-generation climate emulator, which bridges the model hierarchy through effective latent space data assimilation. The resulting bridging model not only inherits the benefits of the operational models, including their high resolution and comprehensive set of variables, but also globally enhances the accuracy through local dynamical and statistical improvements provided by the simpler models.  Our developed latent space technique identifies the dominant nonlinear features, which facilitates effective communication between models. The machine learning representation of the bridging model significantly enhances simulation efficiency, providing high-quality datasets and advancing the quantification of uncertainty in extreme events. The framework is applied to improve the performance of CMIP6 simulations in characterizing ENSO complexity by exploiting simpler yet statistically accurate models. 

Archived semesters


Return to the Applied Mathematics Group Page

Retrieved from "https://wiki.math.wisc.edu/index.php?title=Applied/ACMS&oldid=29023"