SZ4D
Machine Learning & Artificial Intelligence
Virtual Workshop
August 2-4, 2023 (9am-11:30am PDT, 12-2:30pm ET-Chile Time)
Workshop Objectives
To get community input, buy-in, and feedback regarding what AI/ML approaches are most promising, and what specific steps are needed, to achieve SZ4D goals: 1) Connect useful techniques across domains to enable multi-disciplinary science, 2) Determine and prioritize focus areas for future proposals; and 3) Understand community needs for ML/AI training and community software.
Motivation from the SZ4D Implementation Plan
“Given that the initial planning phase of the MCS RCN was focused primarily on physics-based modeling, it will be important to carry out similar efforts for data-driven computational science applications like ML in the next phase of SZ4D Workshops and other community-building activities are needed to identify community needs, opportunities for open-source software development, and training and educational activities.”
The SZ4D Virtual Workshop on Machine Learning and Artificial Intelligence (ML/AI) will take place August 2-4, 2023, with three sessions focused on different themes. Each session will have three invited speakers (1hr), focused breakouts (1hr), & report back/synthesis (0.5hr). More details about each session and link t register in advance below.
Organizers
Daniel Trugman (University of Nevada Reno)
Tushar Mittal (Penn State)
Xuesong Ding (UT Austin)
Topics and Focus Areas
-
Detection of hidden patterns in geoscience data
-
Dimensionality reduction and clustering techniques
-
Methods enabling analysis and interpretation of large datasets
-
Denoising of time series
-
Transient and anomaly detection
-
Edge computing
Confirmed Invited Speakers
​
-
Weiqiang Zhu (UC Berkeley) - FEC
-
Evan Goldstein (UNC Greensboro) - LS
-
Matthew Head (U. Illinois) - MDE
​
​
Presentations
Session 1 Introduction | Daniel Trugman (University of Nevada Reno)
Before I train the model… : Adventures in developing machine learning systems for coastal geomorphology | Evan Goldstein (UNC Greensboro)
Deep learning for earthquake monitoring | Weiqiang Zhu (UC Berkeley)
Machine Learning and the Mogi model: Improving the efficiency of ensemble-based methods for volcano deformation analyses | Matthew Head (U. Illinois)
Session 2
August 3, 9am-11:30am PDT
Theme: Making Predictions with ML/AI
Topics and Focus Areas
-
Forecasting / time series datasets
-
Data assimilation
-
Causality and causal inference
-
Transfer learning across domains/regions
-
Generative models
-
Optimal experimental design
Confirmed Invited Speakers
​
-
Christelle Wauthier (Penn State) - MDE
-
Diego Melgar (University of Oregon) - FEC
-
Chaopeng Shen (Penn State) - LS
Presentations
Session 2 Introduction | Tushar Mittal (Penn State)
Differentiable modelling to unify machine learning and physical models for hydrology and geosciences | Chaopeng Shen (Penn State)
Signal extraction and characterization from geodetic datasets using AI approaches | Christelle Wauthier (Penn State)
Early warning for great earthquakes from characterization of crustal deformation patterns with deep learning | Diego Melgar (University of Oregon)
Session 3
August 4, 9am-11:30am PDT
Theme: Facilitating Process-based Modeling with ML/AI
Topics and Focus Areas
-
Accelerating simulations/calculations with ML/AI
-
Reduced order models with ML/AI
-
Physics-informed ML
-
Bridging scales in space and time
-
Data-driven process or subgrid representation
-
Interpretable AI for geoscience
Confirmed Invited Speakers
​
-
Zongyi Li (Caltech) - LS/MDE/FEC
-
Qingkai Kong (LLNL) - FEC
-
Hamed O’Ghaffari (MIT) - MDE/FEC
Presentations
Session 3 Introduction | Daniel Trugman (University of Nevada Reno)
Rapid 3D Seismic Waveform Modeling using Fourier Neural Operators | Qingkai Kong (Lawrence Livermore National Laboratory)
Geometry-Informed Neural Operator for Large-Scale 3D PDEs | Zongyi Li (Caltech)
Bridging length-time scales in a brittle-ductile process: Evolution of “defects” in fast-slow time space | Hamed O’Ghaffari (MIT)