GenPhys: From Physical Processes to Generative Models

Explore the intersection of physics and generative models in this comprehensive talk by Ziming Liu from Valence Labs. Delve into the concept of Generative Models from Physical Processes (GenPhys), which translates partial differential equations describing physical phenomena into novel generative models. Learn how GenPhys encompasses existing models like diffusion models and Poisson flow generative models while introducing new possibilities such as "Yukawa Generative Models." Discover the potential and limitations of various physical processes in generating models, including the wave equation and Schrödinger equation. Gain insights into the expanding design space of generative models through the lens of physics, covering topics such as universal converters, diffusion, Poisson equations, and dispersion relations. Engage with the Q&A session to deepen your understanding of this innovative approach to generative modeling.

- Intro
- Can physics help build generative models?
- Is there a universal converter from physics to generative models?
- Examples: diffusion, poisson, and wave equation
- Dispersion relations
- Q+A