2026 Conference on Physics and AI: Rose Yu

2026 Conference on Physics and AI: Rose Yu

by Rose Yu

Formal & Physical Sciences AI — Models & Research UYQArtificial intelligence (AI)UYQMMachine learning
🎙 Rose Yu 👥 34K 📅 June 30, 2026 ⏱ 34 min 👁 194 📄 conference talk 🧭 2026-08-03
Available in: English (current) Français

Keywords

AI for sciencescience for AIequivariant networkssymmetryturbulenceclimate modelingNoether's theoremmode connectivityphysics-informed diffusiongenerative models

Summary

Rose Yu, a professor at UC San Diego, presents a talk at the 2026 Conference on Physics and AI, organized by Stanford’s Center for Decoding the Universe. She discusses the entanglement of AI and physics, arguing that they form a single system rather than two separate fields. She first covers AI for science, showcasing methods that incorporate physical inductive biases such as multi-scale dynamics and symmetry into neural networks. Examples include TFNet for turbulence modeling, equivariant networks that adapt to symmetry breaking, and adversarial symmetry discovery that rediscovered Lorentz symmetry from LHC data. She also discusses discovering latent coordinates for reaction-diffusion systems and a physics-informed diffusion model that accelerates climate simulations by 25 times. Then she shifts to science for AI, using physics principles to improve AI itself. She introduces teleportation, an optimizer that exploits symmetry in loss landscapes, and applies Noether’s theorem to deep learning, finding conserved quantities during training and analyzing mode connectivity. She concludes by emphasizing the deep entanglement of AI and physics, suggesting that future progress lies at their intersection.

174 words

Critical Evaluation

The talk provides a comprehensive overview of the intersection of AI and physics, highlighting both directions: AI for science and science for AI. Rose Yu demonstrates a strong command of the subject, presenting concrete examples from her research group that illustrate the benefits of incorporating physical principles into AI models. The argument for entanglement is compelling, as she shows how physics can provide inductive biases that improve AI’s performance in scientific domains, while AI can help discover physical laws and accelerate simulations. The technical depth is high, with references to specific methods like equivariant neural networks, adversarial symmetry discovery, and physics-informed diffusion models. The talk is well-structured, moving from familiar AI-for-science applications to more novel science-for-AI ideas, such as teleportation and Noether’s theorem for deep learning. However, the talk is a conference presentation, so it lacks the rigor of a peer-reviewed paper; some claims are presented without detailed evidence, and the audience is assumed to have a background in both fields. The sources cited are primarily the speaker’s own work and the conference itself, which is appropriate for this format. The title accurately reflects the content. Overall, the talk is informative and thought-provoking, offering valuable insights into the symbiotic relationship between physics and AI.

204 words

Title / Content Match

The title accurately reflects the content: a conference talk on the intersection of physics and AI.

Quality & Reliability

8/10

Talk by a recognized academic (professor at UCSD) presenting research results, with references to published work and conference context. The content is technical and appears rigorous, though not peer-reviewed in this format.

Key Moments

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

Cited Sources

Concurring Sources

Contribution & Novelties

The talk presents several novel contributions from Rose Yu’s group, including TFNet for turbulence modeling, relaxed equivariant models that adapt to symmetry breaking, adversarial symmetry discovery, and a physics-informed diffusion model for climate simulation. It also introduces teleportation, an optimizer that exploits symmetry, and applies Noether’s theorem to deep learning to find conserved quantities and analyze mode connectivity. These contributions advance both AI for science and science for AI.

Pour aller plus loin :

  • Equivariant neural networks — Overview of equivariant networks, relevant to the symmetry-based approaches discussed.
  • Noether’s theorem — Fundamental physics principle applied to deep learning in the talk.
  • Physics-informed neural networks — Related approach for incorporating physics into AI models.
  • Diffusion models — Background on the generative models used for climate simulation.

125 words

Radar Profile

The radar profile shows high scores in technical level and information quality, with slightly lower scores in quantity and reliability, reflecting the depth and specificity of the talk.

Reliability 8/10