Stephen Becker - Some new bi-fidelity methods for UQ and optimization - IPAM at UCLA

Value of the Information & Strength of the Argument

The talk provides valuable insights into bi-fidelity methods, offering concrete algorithmic contributions. The argumentation is solid, with clear problem formulations and mathematical derivations. The UQ method is well-motivated and validated with numerical experiments. The optimization methods are presented with theoretical justifications and practical considerations. The speaker effectively argues for the benefits of using low-fidelity models to reduce computational cost.

Scientific Rigor, Source Quality, Title Accuracy

The presentation is scientifically rigorous, with references to published works (Cheng et al., CMAME ‘24; Cheng et al., TMLR ‘25; Rummel et al., arxiv.org/abs/2604.10089). The methods are based on established principles (VAEs, quasi-Newton methods). The title accurately reflects the content. The talk is part of a workshop, indicating peer review by the community. The description provides a link to the workshop page for further context.

The title accurately reflects the content, which covers new bi-fidelity methods for uncertainty quantification and optimization.

8/10

Presentation of original research with clear methodology, references to peer-reviewed publications, and rigorous mathematical derivations. However, the talk is a workshop presentation and not a peer-reviewed publication itself, and some details are omitted for brevity.

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

• 00:00Introduction and overview of talk topics.
• 01:00Introduction to forward UQ and high-dimensional quantity of interest.
• 03:00Example of a composite beam and bi-fidelity approach using Euler-Bernoulli formula.
• 05:00Explanation of autoencoders and variational autoencoders.
• 08:00Bi-fidelity VAE method: training on low-fidelity samples and mapping to high-fidelity latent space.
• 10:00Results on composite beam problem showing improvement over standard VAE.
• 12:00Transition to optimization: Stokes flow with Janus particles.
• 15:00Formulation of linear complementarity problem for contact forces.
• 18:00Projected gradient descent and quasi-Newton methods for solving LCP.
• 20:00Bi-fidelity quasi-Newton method using low-fidelity Hessian approximation.

• IPAM Workshop: Multi-Fidelity Methods to Enable Robust Optimization and Real-Time Control of Fusion Processes — Workshop page providing context and additional materials.

• Cheng et al., CMAME '24 (referenced in talk) — Reference to a paper on bi-fidelity VAE for UQ.
• Cheng et al., TMLR '25 (referenced in talk) — Reference to a paper on stochastic subspace descent with low-fidelity line search.
• Rummel et al., arxiv.org/abs/2604.10089 (referenced in talk) — Reference to a paper on bi-fidelity quasi-Newton methods for Stokes flow.