Physics-Informed Laplace Neural Operators || ML linear algebra algorithms || March 13, 2026

Value of the Information & Strength of the Argument

The first talk provides a clear motivation for physics-informed neural operators, addressing limitations of purely data-driven approaches in small-data and out-of-distribution settings. The argumentation is solid, with conceptual illustrations and experimental results supporting the claims. The second talk presents novel frameworks for learning linear algebra algorithms, with empirical evidence of improved performance. Both talks are technically rigorous and offer valuable insights into their respective fields.

Scientific Rigor, Source Quality, Title Accuracy

The seminar is scientifically rigorous, with references to prior work such as DeepONet, FNO, and physics-informed methods. The sources cited are relevant and credible. The title accurately reflects the content, covering both talks. The presentation is well-structured and the technical details are appropriately detailed.

The title accurately reflects the two main talks: physics-informed neural operators and ML linear algebra algorithms.

8/10

The seminar presents original research from two established researchers, with detailed technical content and references to prior work. The claims are supported by experimental results, though not peer-reviewed in this format.

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

• 00:00Introduction to the seminar and first speaker.
• 02:00Background on PINs vs operator learning.
• 10:00Introduction of Laplace Neural Operator and its limitations.
• 20:00Proposal of Physics-Informed Laplace Neural Operator (PILNO).
• 30:00Explanation of virtual inputs and their role in data efficiency.
• 40:00Temporal causality weighting for time-dependent PDEs.
• 50:00Numerical results on Burgers' equation and other benchmarks.
• 1:00:00Conclusion of first talk and Q&A.
• 1:10:00Introduction of second speaker, Michael Mahoney.
• 1:20:00Overview of PRISM framework for matrix functions.
• 1:30:00AutoSpec: neural network for discovering spectral algorithms.
• 1:40:00Applications and implications for RandBLAS/RandLAPACK.

• Laplace Neural Operator — Referenced as the base architecture for PILNO.
• Physics-Informed DeepONet — Mentioned as prior work in physics-informed operator learning.
• Physics-Informed AFNO — Mentioned as prior work in physics-informed operator learning.
• PRISM — Introduced in the second talk as a framework for accelerating matrix function computations.
• AutoSpec — Introduced in the second talk as a neural network framework for discovering spectral algorithms.

• DeepONet — Referenced as a foundational operator learning method.
• Fourier Neural Operator — Referenced as a comparison and basis for the Laplace Neural Operator.