New Probabilistic Foundations of Uncertainty Quantification and Machine Learning: How to Model What We Don’t Know
Dr. Frank Günther
Director of Analysis & Simulations at Knorr-Bremse Rail Systems
Member of the NAFEMS Stochastics Working Group
Dr. Frank Günther is the Director of Virtual Testing and Simulations at Knorr-Bremse SfS, the world market leader for railway brake systems, as well as many other railway technologies. In this role he leads several diverse, international teams responsible for all aspects of CAE and numerical simulation. Dr. Günther has many years of experience both as a leader and an expert in numerical analysis and simulation technology. His experience has led him to become a proponent of Virtual Testing, Simulation Governance, Physics-Informed Machine Learning / AI, and Uncertainty Quantification. Before joining Knorr-Bremse, Dr. Günther worked as a senior crash simulation expert and project leader at Daimler AG in Stuttgart, Germany. He started his career by obtaining a PhD in the Theoretical and Applied Mechanics Group of Northwestern University in Evanston, IL (USA).
- Bayesian and Frequentist concepts of probability,
- Aleatory and epistemic uncertainty,
- Tools for statistical machine learning,
- Statistical modeling: data generating model and prior predictive simulation,
- Bayes’s rule as the foundation for statistical simulation,
- (Machine) Learning from data and posterior predictive simulation.
- Uncertainty quantification as an important element of Simulation Governance,
- Numerical solution of statistical models: Kalman filters, particle filters, Markov Chain Monte Carlo,
- Example: Using statistical machine learning to predict fatigue test outcomes,
- Example: Using statistical machine learning to enhance faulty and incomplete measurement data through sensor fusion.
A full set of notes in PDF format as well as simulation case files in ZIP format will be available for download for the seminar attendees. All the above will be explained in a purely practical manner without much theory. Most explanations will be done on simple examples attendants will be welcomed to try for themselves.
- You will learn about the basics of Uncertainty Quantification and Statistical Machine Learning starting from first principles such as concepts of probability and the application of Bayes’s Rule.
- You will see how statistical simulations can be run and evaluated using freely available software such as R, Stan, and RStudio.
- You will learn about application use cases of Statistical Machine Learning in industry.
What questions will this course answer?
- What is probability and how can I use it to assess and enhance the predictive power of simulations?
- How can I use statistical modeling to learn from data?
- How can statistical simulation bridge the gap between traditional simulation and “physics-agnostic” machine learning?
- How can I make predictions and decisions in the face of uncertainty?
- Everyone is welcome to attend. The seminar will be most interesting for
- Simulation engineers that would like to understand the predictive power of their models by quantifying uncertainty.
- Engineers that would like to broaden their understanding of potential applications of statistical machine learning in their field.
- Anyone interested in the practical application of physical modeling, machine learning, and data science.