TrAC Scientific Machine Learning
The Scientific Machine Learning (SciML) digital badge focuses on theory, implementation, and limitations of scientific machine learning models including physics-informed neural networks (PINNs) and state-of-the-art neural operators. The course is designed to develop skills in tackling partial differential equations using machine learning. The SciML course is offered by Iowa State University's Translational AI Center (TrAC) and is part of a larger Advanced AI Techniques pathway program and intended for audiences within the spectrum of the software and technology industry, including software engineers, data scientists, data engineers, data analysts, research scientists, and software developers.
Skills / Knowledge
- Artificial Intelligence
- Machine Learning
- PyTorch
- Artificial Neural Networks
- Partial Differential Equation
- Problem-solving
- Physics Informed Neural Networks
Earning Criteria
Required
The Scientific Machine Learning (SciML) Badge is earned after successful completion of a 4-week, asynchronous, self-paced online course consisting of 4 modules. The 4 modules cover essential topics including physics-informed neural networks (PINNs) and state-of-the-art neural operators (PyTorch).
This course offers a blend of hands-on activities, assignments, video lectures and tutorials.
Learning Outcomes:
Discretize PDES and solve using numerical methods (finite difference methods)
Implement heat equation using finite difference methods in python
Assess the performance of PINNs and compare it to standard numerical methods
Identify common challenges in Physics-Informed Neural Networks (PINNs), such as hard convergence and weighting of BCs and ICs and implement strategies to overcome them
Implement PINNs for 1D steady state heat equation using PyTorch
Explain the theory of operator learning, including the concept of continuous-discrete equivalence
Analyze Fourier Neural Operators, their implementation and limitations (such as aliasing errors)
Implement Fourier Neural Operator to solve PDEs in a data-driven approach
Analyze deep operator networks and their architecture
Implement DeepONet to solve PDEs in a data-driven approach
Assessment:
Participants will be assessed on:
Engagement with each module
Two coding assignments centered on a simple physics-informed neural network and utilizing state-of-the-art neural operators (FNO, DeepONet) to learn PDEs from data
2 quizzes on design and to debug SciML models
About TrAC
The Translational AI Center will break down disciplinary silos to bring together core Iowa State artificial intelligence researchers and subject matter experts interested in applying new technologies to their work. The center will initially focus on conducting core artificial intelligence research, as well as pursuing five application areas of artificial intelligence:
Materials design and manufacturing
Biology, healthcare, and quality of life
Autonomy, intelligent transportation, and smart infrastructure
Food, energy, and water
Ethics, fairness, and adoption.
In addition to serving as a scientific hub for translational artificial intelligence, the center will organize research seminars, host workshops, training, and onboarding programs, offer seed funding for research projects, and serve as an intermediary between private industry partners seeking research services and appropriate university faculty.