Karen Willcox - Learning Structure-exploiting Reduced Models with Operator Inference

Value of the Information & Strength of the Argument

The talk provides a high-value overview of Operator Inference, a method with significant potential for accelerating simulations in engineering and science. The argumentation is solid, grounded in a decade of research and peer-reviewed publications. Willcox clearly explains the motivation, methodology, and advantages over traditional intrusive methods, while also acknowledging limitations and areas for future work. The presentation is well-structured and persuasive, making a strong case for the adoption of data-driven reduced-order modeling.

Scientific Rigor, Source Quality, Title Accuracy

The talk demonstrates high scientific rigor, with references to specific papers and collaborations. The sources cited are credible, including the original Operator Inference paper (Peherstorfer & Willcox, 2016) and subsequent works. The title accurately reflects the content, focusing on learning structure-exploiting reduced models. The presentation is based on established theory and numerical experiments, with clear explanations of the mathematical foundations. The talk is part of an IPAM workshop, adding to its credibility.

The title accurately reflects the content, focusing on learning reduced models with operator inference and exploiting structure.

9/10

Presentation by a leading expert in reduced-order modeling, based on a decade of peer-reviewed research, with clear methodology and references to specific papers.

Markers derived by PSI from the transcript: the creator did not define chapters.

• 02:00Introduction to reduced-order models and motivation for their use in outer-loop applications.
• 05:00Comparison of reduced-order modeling with machine learning, highlighting differences in philosophy and approach.
• 10:00Explanation of projection-based model reduction and the concept of preserving physics structure.
• 15:00Introduction to Operator Inference and its formulation as a linear least squares problem.
• 20:00Discussion of the four ingredients: physics, projection theory, inverse theory, and numerical linear algebra.
• 25:00Detailed derivation of the reduced-order model for linear and quadratic systems.
• 30:00Overview of the Operator Inference workflow: snapshots, POD basis, projection, and least squares.
• 35:00Discussion of recent advances: nested formulations and block-structured formulations for multiphysics systems.
• 40:00Examples and applications, including rotating detonation rocket engine simulations.
• 45:00Future directions and potential for broader adoption of Operator Inference.

• IPAM Workshop: Learning Models from Data for Multi-Fidelity Fusion Plasma Physics — Workshop where the talk was presented, providing context and related resources.

• IPAM Workshop: Learning Models from Data for Multi-Fidelity Fusion Plasma Physics — Workshop context supporting the relevance of the talk.